IoT Vulnerability Assessment for Sustainable Computing: Threats, Current Solutions, and Open Challenges

Over the last few decades, sustainable computing has been widely used in areas like social computing, artificial intelligence-based agent systems, mobile computing, and Internet of Things (IoT). There are social, economic, and commercial impacts of IoT on human lives. However, IoT nodes are generally power-constrained with data transmission using an open channel, i.e., Internet which opens the gates for various types of attacks on them. In this context, several efforts are initiated to deal with the evolving security issues in IoT systems and make them self-sufficient to harvest energy for smooth functioning. Motivated by these facts, in this paper, we explore the evolving vulnerabilities in IoT devices. We provide a state-of-the-art survey that addresses multiple dimensions of the IoT realm. Moreover, we provide a general overview of IoT, Sustainable IoT, its architecture, and the Internet Engineering Task Force (IETF) protocol suite. Subsequently, we explore the open-source tools and datasets for the proliferation in research and growth of IoT. A detailed taxonomy of attacks associated with various vulnerabilities is also presented in the text. Then we have specifically focused on the IoT Vulnerability Assessment techniques followed by a case study on sustainability of Smart Agriculture. Finally, this paper outlines the emerging challenges related to IoT and its sustainability, and opening the doors for the beginners to start research in this promising area.


I. INTRODUCTION
The way Internet has reformed the world, we can hardly envisage our lives without it. We are living in the era where various objects across the globe are connected to the Internet. These objects are uniquely identifiable and can sense, actuate, and communicate without human intercession [1]. The journey of objects to smart objects is based on the amalgamation of the Internet with emanating technologies like cloud computing, embedded sensors, Wireless Sensor Networks (WSN), middleware, and Radio-frequency identification (RFID) [2]. This amalgam seeded the word IoT, a wired /wireless network of uniquely identifiable connected The associate editor coordinating the review of this manuscript and approving it for publication was Zhenyu Zhou . things that are capable of processing data and communicating with each other with or without human intervention [3]. The IoT has eased the process to monitor and control the environments by linking the physical world with the web [4].
IoT services have a major impact on the lives of people. The people-centric solutions, like IoT assistance, allow the disabled people to enjoy independence and participation in their social life [5]. Moreover, the IoT solutions assist in in-home rehabilitation for physical therapy [6]. In contrast, the Autism Glass helps autistic children to make out facial emotions of people and thus aids in social interactions [7], [8]. Additionally, IoT solutions aids in minimizing hazardous situations. For example, IoT has made the dangerous tasks of mining safer and efficient like self-driving autonomous mining tools keep the workers apart VOLUME 8, 2020 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ from hazardous zones. The location and proximity sensors also aid in the same [9]. There are several IoT sensors such as smoke, toxic gas, temperature when couple with warning systems prevent environmental disasters. These sensors can also keep a check on chemical leaks in water bodies [10]. A lot of case studies have also been reported by various research institutes in collaboration, to show the influence of IoT on natural resources [11]. The certain services provided by the IoT systems come infusion with significant security flaws. Manufacturers overlook the security considerations and produce devices that could be easily exploited. It appeared that 70 percent of Internet-connected devices are vulnerable to cyber-threats [12]. Moreover, as per the studies by the end of 2020, 25 percent of industrial attacks will be due to compromised IoT devices [13]. This severity can be seen from the number of cyber-attacks like-Mirai (2016), Hajime botnet (2016), Persirai (2017), and BrikerBot (2017) launched successfully by exploited IoT devices [14], [15]. Furthermore, privacy is also hindered. IoT based baby monitors and IoT toys [16] are played with by hackers to get sensitive information like video streaming of baby monitors [17], voice recordings of parents, and their kids (in millions), emails, passwords, etc. Easy reprogramming of IoT Device firmware is an add-on for the adversary [18]. Above all, IoT could be a severe threat to flesh and blood. The US Food and Drug Administration also confirmed the risks allied with the reconfiguration of implantable devices and their unauthorized access [19]. All this raises the alarm to take security and privacy issues as a serious matter of concern for sustainable IoT [20], [21].
On similar lines, the energy requirement for IoT devices and their communication plays a crucial role leading to sustainable IoT. Over the past decade, the digital environment and smart devices have increased energy consumption to an alarming level. The renewable sources of energy must be incorporated in energy harvesting (EH) to power widespread IoT sensors [22], [23]. Because batteries of IoT sensors have limited lifetime and its impossible to frequently charge or replace them as they need to run for an extended period of time. For example, in body sensor networks, the EH-enabled sensors along with continuously monitoring the patient can harvest the energy from the patient's body [24] or environment, like thermal energy, kinetic energy,solar energy, and radio frequency signals [25]. With energy harvesting, another promising solution to address this challenge is an efficient data transmission scheme [26]. It is found that 80% of the sensor's energy is consumed during data transmission. Moreover, EH chips are also being attacked by malicious Trojans destroying sensors and thus leading to DoS attacks. Hence, both the factors security and energy-efficiency define sustainable IoT. However, the two are the conflicting challenges for the growth and operation of IoT [27]. Because IoT nodes being power constrained need lightweight energy-efficient security mechanisms [28]. In this article, we will cover the security as a challenge for the sustenance of IoT. Specifically, the vulnerabilities in an IoT system, that serves as the doorway to numerous threats and posing a significant risk to sustainable IoT. The exponential growth and resource-constrained nature of IoT devices challenge confronting various IoT security issues. FIGURE 1 shows the estimated IoT enabled devices (internet-connected) will be 80 billion by the year 2030. Although several security mechanisms exist in literature to enhance the security of IoT. The existing IoT solutions either impose computational load on IoT devices or are so lightweight that they could easily be bypassed. The higher computational load will lead to early battery-depletion of smart nodes. With self energy harvesting capacity [29], the node will be more efficient to secure and sustain itself in this connected era. Moreover, to meet the long-term power budget of these constrained IoT nodes, the researchers across the globe have given their energy-efficient solutions to meet the growing challenges like security, privacy, and interoperability. For example, being resource-constrained, the IoT nodes offload their computational overhead to the edge-servers through specific channels in an energy-efficient way [30]. On similar lines, the recent works provide secure and energy-efficient solutions [31], [32]. One of such solutions is blockchain-based secure and efficient energy trading from vehicle-to-grid and the other way in Energy Internet [33].

A. SCOPE OF THIS SURVEY
The IoT architecture, protocols, growing technologies, IoT attacks, and threats have been widely studied in the reviewed literature. However, no comprehensive survey exists which has covered the IoT vulnerabilities and their assessment in context to sustainable computing. For example, Gupta et al. [1] have put together in their survey the historical background of the IoT, methodically studied the architecture of IoT, and variant nature of challenges it can come across. They have also weighed up permissive technologies like RFID and WSN, along with their key issues and existing solutions to grapple with. Similarly, Atzori et al. [34] explored IoT in multiple contexts, discussed enabling technologies and their impacts in everyday-life. We have examined many such correlated surveys to find their contributions and illustrate how the present study progresses the state-of-the-art in terms of IoT security.
Sicari et al. [35] reviewed the existing state-of-the-art solutions in the field of IoT security. The authors also explored the proposals on security middlewares and solutions for mobile devices. Some ongoing international projects are also studied. Finally, they have given the future directions. One being the need of unified vision for assurance of security requirements in different environments. In contrast, Granjal et al. [36] provide deep insight for communication protocols in IoT, such as IEEE 802.15.4, IEEE802.15.4.e, 6LowPAN, RPL, and Constrained Application Protocol (CoAP). They also explored the security provided by these protocols in the communication stack of IoT. Moreover, research challenges and proposals for security against packet fragmentation, key management, solutions against internal attacks, and compressed security headers for the 6LoWPAN adaptation layer put forward to secure communications availing the IoT technologies forging the protocol stack.
Samaila et al. [37] performed survey covers to multiple security concerns such as system model, threat model. Further, they thoroughly explored nine application domains with their different models, associated assets, and security requirements. The authors also discussed solutions based on cryptographic primitives, authentication mechanisms, hardware, and specific application domains. The paper highlights the current IoT security mechanisms and open issues that need to be addressed. On similar lines, Roman et al. [38] analyzed the features and the security challenges in centralized and distributed IoT to cognize their sustainability in IoT. Additionally, Zhang et al. [39] discovered five weak areas about IoT security by mapping real IoT incidents with the existing security solutions. They are implementation loopholes, inadequate authentication, excessively privileged applications, environmental mistrust, and LAN mistrust. Moreover, the authors provided their dataset and statistics online.
In addition to this, Alaba et al. [40] proposed the taxonomy of IoT security, in terms of application, architecture, and communication. The authors also discussed numerous attacks launched by exploiting threats and vulnerabilities in IoT [21]. Moreover, some emerging IoT challenges related to trust, security, and infrastructure were talked about. In this way, the authors reasoned that the diversity of resource-constrained IoT devices hampers the scalability of promising security solutions. In another work, Mosenia and Jha [41] presented several attack scenarios, and their potential mitigation approaches for the Cisco 7-layer reference model [42]. The authors emphasized the significance of using a proactive approach to secure the IoT environment. They analyzed the vulnerabilities and provided necessary countermeasures for edge nodes, communication, and edge computing in an IoT system. Furthermore, they briefly described the IoT reference models, applications of IoT, and the attack vectors. Finally, they discussed two emerging security challenges -Unexpected usages of data, Exponential rise in the frequency of weak links.
Neshenko et al. [10] centered their work on emerging IoT security vulnerabilities. The survey presented the unique taxonomy on IoT vulnerabilities, which includes layer wise vulnerabilities, their security impact, their attack vectors, remediation strategies, and situational awareness capabilities. Furthermore, they proposed a data-driven approach for empirical assessment of IoT maliciousness. In addition, the authors drew insightful findings and inferences in various sections of the survey. On similar lines, Mahbub [43] and Srivastava et al. [44] presented the comprehensive work on growing security challenges in terms of vulnerabilities and threats. In another notable work, Makhdoom et al. [12] highlighted threats in context to IoT architecture and had given due diligence on the taxonomy of malware attacks with their attack approach. The authors also discussed the DDoS attack strategy by making a botnet of IoT motes, followed by needed security measures. The authors have given a comprehensive set of security guidelines grounded on industry best practices to apply minimum security standards in an IoT system. In the end, some open challenges, the lessons learned, and pitfalls are included within. TABLE 1 shows the relative comparison of the proposed survey with state-of-the-art surveys. In this table, the readers could easily identify already available contributions in the state-of-the-art. They mainly centered their surveys around IoT architectures, Protocols for resource-constrained devices, enabling technologies, IoT attacks, threat modeling, and countermeasure strategies. From studied start-of-theart, we noticed there are few surveys, which precisely emphasize on the growing IoT vulnerabilities. Furthermore, these surveys provide insight into IoT security threats and proposed solutions only from a general perspective. None of them addresses the recent trend of Machine Learning (ML) and other vulnerability assessment techniques and IoT security [45], [46]. Addressing these recent trends diverged the research towards the key tasks of discovering a pattern in enormous data, detecting outliers, extracting features for vulnerability detection, and predicting performance estimation metrics for IoT enabled systems using ML [47]. The proposed survey covers these research gaps and focuses mainly on emerging IoT vulnerabilities and various vulnerability assessment techniques to secure IoT devices for sustainable IoT [48].

B. RESEARCH CONTRIBUTIONS
Following are the contributions of this paper.
• We have presented a taxonomy that focuses on energy-efficiency and security for sustainable IoT.
• We have highlighted the benefits of the growing usage of techniques for the IoT vulnerability assessment such as machine learning, honeypots, fuzzy techniques, and penetration testing tools. • A case study on Sustainable Smart Agriculture has been presented.
• Finally, various open issues and future recommendations to ensure secure and sustainable IoT infrastructure for the end-users have been given.

C. METHODS AND MATERIALS
The proper methodology is adopted to conduct this survey in an appropriate manner to give a detailed analysis of two critical pillars, security, and energy for sustainable IoT.  The rest of the paper is organised as follows. Section II provides the overview of IoT, which includes sustainable IoT, the architecture of IoT, its protocol suite, and open source tools for IoT. IoT security vulnerabilities are discussed in Section III. The taxonomy of IoT vulnerability assessment techniques is discussed in Section IV, followed by a case study in Section V. In Section VI, we present the findings of the paper and finally, Section VII concludes the paper.

II. BACKGROUND AND PRELIMINARIES
This section focuses on the background and importance of security in IoT. This section is bifurcated into three subsections. Firstly, we discuss IoT architecture. Secondly, we discuss the protocol suite for IoT. In the third subsection, we focused on open source tools and datasets. Kevin Ashton firstly proposed IoT in 1999, and he referred the IoT as ''uniquely identifiable interoperable connected objects with RFID technology'' [56]. Around the early 1980s, the intelligent/smart coke machine was connected to the Internet to take the invoice of the list of coke drinks available, and this brought the concept of interlinking among the smart things [1]. The IoT, being an emanating world network of uniquely referable computing devices within the existing Internet infrastructure, is renewing our lives and the way we work by proliferating the connectivity of people and things to an unimaginable extent. In addition to diverse and profound applications of IoT, the rising security issues cause inestimable consequences [57]. TABLE 3 illustrates the main milestones allied with the evolution of IoT since 1999.

A. SUSTAINABLE IoT
Though IoT has become an integral part of our lives, there are a huge number of devices which have no mechanisms for energy harvesting and security. These two factors must be prerequisites at the design phase and all the aspects of their life-cycle (sensors) must be addressed right from deployment to their disposal. Thus, for sustainable IoT, energy sustainability and security sustainability are two critical pillars. The sustainable IoT is very well represented in FIGURE 3, where energy sources are used as a supply for IoT end nodes and security solutions preventing the malfunctioning of an IoT system. Being power-constrained, IoT end nodes are the weakest point in an end-to-end system. The energy efficiency solutions in terms of power consumption and data transmission have become the present need for sustainable IoT. As IoT revolves around data, the fate of IoT depends upon the security and privacy of the same [59]. The recent security breaches depict that even resource-constrained IoT end nodes with limited functionality induce substantial risk to the whole system. This is because of the connected nature of the IoT devices which provides a large attack surface, forming numerous attack points for the adversaries.  • Energy Sustainability: The overpowering IoT services infused in our lives raises the concern for the power-constrained IoT nodes for sustainable IoT. The mass deployment of IoT sensors and actuators in several sectors require a continuous power supply for a prolonged period. Because as the size of these IoT nodes being sensors or actuators is getting small, their battery size is also reducing. Thus, stores the reduced amount of energy in these end nodes. Also, the lifetime of the battery is considerably smaller than the electronics. Moreover, the growing trend is to add more functionality to these power-constrained devices, which generally works in a wireless mode. This is a call for considerable progress in energy efficiency for both communication and computation in power-constrained IoT nodes for their sustenance [28]. Energy harvesting is one of the solutions and incorporated in several IoT application domains for wearables, bridges, road sensors, dams, mines, and drones. In many EH schemes, the energy from the surrounding environment aids in powering sensors and communication technologies.
The radio frequency signals [25], solar [27], wind, water, human body [24], and piezoelectric are some of the common energy sources in EH schemes [22]. The maximum power point tracking schemes aid in extracting maximum energy from the input to boost power efficiency [60]. The inductor-less design for solar energy management [61] and several other energy management systems are designed in the literature. The bandgap-based output controller is used for output regulation with EH [62]. For boosting low input energy DC-DC converters are used as charge pumps [63]. Several wi-fi based EH schemes with efficient data transmission like CoWiFi [64] and context-awareness schemes are also designed. Though lot many efforts are made in improving the energy-efficiency in power-constrained IoT systems, but the pace does not match with the emerging IoT dependence/services.
• Security Sustainability: IoT being linked with real physical world phenomenon such as healthcare, agriculture, grids, weather, and taking decisions based on sensing and monitoring, necessitates the special concern in security [65], [66]. For sustainable IoT, data and device security both need to be taken into consideration. The data security mainly covers the integrity and confidentiality of data, whereas devices need to be protected from stealthy attacks. The common IoT security vulnerabilities hindering the sustainability of IoT remain unnoticed throughout the development and shipment period. Generally, the things which are a part of IoT to provide smart services are the vulnerable things [67]. For example, IoT components with obsolete OS versions, weak hard-coded passwords, insecure firmware updates, improper authentication mechanisms, open debugging ports, and insecure interfaces [68]. Even they impose a significant risk to human lives. As per the reports, more than 70 percent of smart devices are prone to stealthy cyber-attacks [12]. Additionally, in near future, 25% of industrial attacks [69] will be caused by compromised smart devices. This scrupulousness could be understood from industrial cyber-attacks like Stuxnet attack [70], and attack on German steel mill [71]. Thus, adversaries easily exploit resource-constrained IoT devices as other connected devices like laptops, desktops, etc. are protected with stable guarding mechanisms. In this article, the root cause of growing threats namely the security vulnerabilities in an IoT system will be covered. For the sustainable functioning of an IoT system, the balance must be maintained among the interdependent features like energy efficiency, performance, security, and power consumption. The small battery size with a reduced amount of energy lessens the resource availability to secure these power-constrained devices. It is found that with the decrease in resources for security, there is a continuous increase in security requirements of IoT end nodes, pushing the significant research initiatives in lightweight security technologies for constrained devices. The traditional security mechanisms like cryptographic solutions developed for powered devices need more computations and thus consume more power. The state-of-the-art light-weight cryptographic schemes show that Advanced encryption standard and Elliptic curve cryptography are the most preferred one, when compared in terms of limited resources, throughput, chip area, and latency [72].

B. ARCHITECTURE OF IoT
The numbers of IoT framework have been presented by international organizations and working groups namely; International Telecommunication Union [39], Institute of Electrical and Electronics Engineers (IEEE) [40], European Telecommunications Standards Institute [41] and Cisco [42], based on variant nature of requirements of IoT environment. Even so, none of them have been standardized until now. Several research efforts are made to build IoT architecture to meet security requirements. TABLE 4 summarizes the existing IoT architecture and its related security domains. The general IoT architecture given by ITU -Telecommunication Standardization Sector Y.2002 is briefly described in [39]. In this architecture, there are three layers namely Perception Layer, Network Layer, and Application Layer.
Perception Layer: It is the lowest layer in the IoT architecture where the IoT nodes can be RFID readers, RFID tags, QR code, Bluetooth devices, GPS devices, multiple sensors like light, humidity, temperature and so on. These devices could serve different purposes [1], which are as follows. Application Layer: It is the topmost layer of IoT architecture, which provides IoT based services to the users globally by using different devices like laptops, mobiles, and personal digital assistants. It provides an interface through which the user can interact with its system. IoT has a wide range of application domains. These include commercial applications, industrial applications, applications specific to people, and consumer-oriented applications as shown in FIGURE 4.

C. PROTOCOL SUITE FOR IoT
IoT being a realm of resource-constrained motes, cannot rely on TCP/IP protocols such as IPv4, TCP, and Hypertext Transfer Protocol (HTTP). Relying on them may lead to wastage of energy during transmission in the form of voluble metadata, protocol overheads, and non-optimized communication patterns. The working groups of standardization bodies IEEE and IETF have put forward the communication protocols for resource-constrained devices [58]. The formalized protocol stack proposed by the author as shown in FIGURE 5. • Perception Layer IEEE 802. 15 preserving very low duty cycles, it endowed high reliability through time synchronization and channel hopping, using a scheme originally propounded in the form of TSMP [36]. Wireless Hart [84] is also based on this protocol. In TCHP, devices synchronize as per slot frame structure, and a set of slots iterating over time. A schedule is followed by each device that states what to do in every slot. A mote can sleep, receive, or transmit in a particular slot. The mote keeps its radio off in a sleeping slot. For each active slot, the schedule includes the channel offset and the neighbor to whom it gets to transmit or receive. It also defines how the schedule will be executed in the Mac Layer, which may be centralized or distributed. Few modifications are also done to improve the security at the MAC layer by the IEEE 802.15.4e working group [36], [58].  [85]. Moreover, it provides stateless IPv6 header compression, mesh routing, and simplified IPv6 neighbor discovery protocol.
• Network Layer: To develop the IPv6 routing protocol for Low-Power and Lossy Networks, the IETF RoLL working group was created in 2008. By utilizing routing requirements and quantitative metrics for nodes and links, RoLL developed a Routing Protocol for Low-Power And Lossy networks. It is a distance-vector routing protocol, which allows the nodes to exchange distance vectors and root with a controller to create a Destination-Oriented Directed Acyclic Graph. It aids three kinds of traffic flow: multipoint-to-point, pointto-multipoint, and point -to-point [50]. It is dealing with several issues like end-to-end throughput challenge due to the co-existence of multiple applications in one physical network, packet re-ordering, and rises in the cost of DAG creation and maintenance due to multipath routing structure, and effect of duty-cycling on end-wise latency. The number of solutions has been proposed to conquer them like Queue-aware backpressure routing algorithm, opportunistic routing and networking encoding [1], load balancing [86], and adaptive control on duty cycling [87].
• Application Layer: The IETF CORE group has designed CoAP, a protocol for web transfer in a constrained environment. We cannot say CoAP as a wadded form of VOLUME 8, 2020 HTTP [88]; it's just a part of Restful specification, which makes it compatible with constrained environments. CoAP endorses datagram-oriented transport protocols, such as User Datagram Protocol (UDP). CoAP aids reliable transmission over UDP. A messaging layer is responsible for reliability and sequencing, whereas a request/response layer maps requests to responses as well as their semantics. The conspectus of the main features [58] provided by CoAP is as follows. 1) A web protocol specialized in Machine-to-Machine requirements and a constrained environment.
3) It supports unicast and multicast requests by binding UDP with optional reliability. 4) Enables Asynchronous message exchanges and built-in resource discovery. 5) Parsing complexity and Low header overhead. 6) Limited to simple proxy as well as caching capabilities.

D. OPEN SOURCE TOOLS AND DATASETS
There exist many open-source tools, which accelerates the growth of IoT-based applications. Moreover, open-source tools and datasets aid researchers in formulating theories, devising experimental results, and developing system models.  The Arduino is an open-source electronics platform, that helps in developing IoT systems. It consists of a microcontroller which can be programmed with the help of Arduino programming language; can take inputs which can be either a simple text message, light sensed by a sensor, a fingerprint, etc. and can produce outputs like turning on the motor, lighting an LED, sending the text message, etc. Moreover, Arduino Software could be used for the same. All Arduino boards, Arduino programming language, and Arduino Software are open-source, emancipating developers to freely design them and use them as per their individual needs [89]. Eclipse IoT is a working group of companies which run open source community for IoT. More than thirty companies are working together, namely IBM Redhat, Bosch, Kichwa coders, Eurotech, V2com, etc. It provides everything needed to build IoT solutions moving from constrained devices, gateways, cloud platforms, standards, protocols, etc [94]. It also provides IoT open source projects, resources like case studies, white papers, newsletters, and aids virtual IoT Meet up [90], [95].
Beagle Board is a non-profit corporation that makes the masses about the design and uses open-source hardware and software within embedded computing. It also provides a forum to exchange ideas. Moreover, headways towards open-source computing solutions comprising robotics, machine controls, and manufacturing tools like 3D printers. Beagle Boards are fan-less boards with power-efficient Texas processors, even expansible to desktop machines, unaccompanied by bulk, expense, or noise. The open-source designs of these boards are also available for making compatible hardware [91]. An Italian company, IoMote, which provides a range of programmable, Arduino-compatible, IoT Edge Devices so that anything can be connected to cloud easily. It empowers reliable and secure bi-directional communications between millions of IoT devices, using Mymote Cloud software, running on Microsoft Azure [102]. It provides products like X400an IoT Edge Gateway, Arduino-compatible easy to program and appropriate solution for high-end IoT projects that require optimal security and bidirectional real-time communication. Similarly, they have come up with XSense: NB-IoT Wireless Sensors, embedded with a large number of variant possibilities of flexible sensors for air, noise, water, and many more [103]. It provides longer battery life, global coverage, resilient with problems due to walls, and cost-efficient [104].
Arduino Ethernet Shield connects the Arduino board to the Internet with the help of Ethernet library and activates it to communicate across the world [105]. OpenIoT has come up with a platform to design and manage environments containing IoT resources. It also leverages on-demand utilities for IoT systems, for example, sensing-as-a-service [106], [107].
Contiki [108] is also an open-source operating system for IoT that connects tiny low-cost, power-efficient microcontrollers to the Internet. It supports IPv6, IPv4, with low-power wireless standards. It has provided a lightweight flash file system, called Coffee; an optional command-line shell, tailored wireless networking stack called Rime. A set of nightly regression tests are run on a daily basis in the Cooja simulator, for testing the Contiki code [109].
Raspberry Pi [110] is a card-sized affordable computer that could be used for several purposes as for learning programming, IoT projects. It is not entirely open-source, though the software and documentation are. Raspbian is an operating system for Raspberry Pi, based on the Debian distribution of Linux [111]. DeviceHive is a free, highly scalable open-source IoT platform that provides modules for data collection, processing, and analysis, visualization, device management [92], etc. For developing Smart home solutions, various Home Automation Softwares are available like Eclipse Smart Home [93] and The Thing System [112].

III. IoT VULNERABILITIES: CONCEPT AND SECURITY ASPECT
Due to vulnerability breaches and cyber-attacks the security of IoT is in an alarming state [55], [113]. The number of unanticipated vulnerabilities and exploits are reported, that was designed to take advantage of security gaps in systems and deployment configurations. Some of them, like Mirai, BrickerBot, and Hajime, are discussed in the following subsections. Traditionally, security requirements were mainly defined by three properties: confidentiality, integrity, and availability, as coded by CIA Triad. But the security mentioned above properties of CIA Triad turned out to be insufficient in the context of security [114], [115]. A comprehensive list of security requirements known as the IAS-Octave [114] taken as an extension to CIA Triad is summarized in TABLE 6. Also, the transition from the CIA Triad to IAS-Octave is shown in FIGURE 7.
Vulnerability is a kind of a hole or flaw in a system which if left unhandled, could lead to serious threats to the whole system. These security threats could be seen concerning different layers in IoT Architecture, as shown in FIGURE 8. Talking about the sheer number of IoT application domains, which are in no way less affected by these threats, described briefly in TABLE 7. Moreover, IoT manufacturers treat security as an afterthought. Like that, IoT will lose all of its incredible potentials. The vulnerability assessment will play a significant role in protecting IoT devices from these growing threats. In the sequel, under FIGURE 9, we elaborate on the device-based proposed taxonomy of vulnerabilities in IoT devices [116]. a) Physical Security of IoT Nodes: The IoT nodes must be physically hardened to prevent the risks associated with the direct physical access to these unattended devices. The adversary enters the system through USB or some other ports kept open for maintenance or configuration. He can directly access the SD card and other storage medium to get the control over operating system and gain sensitive data like embedded passwords. Thus, the lack of physical hardening can result into attacks like node cloning and side-channel attacks.
• Node Cloning: The clones of the IoT nodes could be made with ease as they remain in unattended surroundings. Moreover, there is no standard VOLUME 8, 2020 mechanism to develop IoT devices with a hardware tamper-proof. By making replicas of IoT nodes, the adversary could launch the number of attacks. He can use the credentials of compromised nodes to have access over the network [12].
• Side-channel attacks: These attacks aim at getting the side channel information about the device performing cryptographic operations. This information includes physical characteristics of a machine while carrying out those operations, i.e., data about the power consumption, processing time, electro-magnetic emissions, and the sounds it produces. Then this information is used to reverse engineer the cryptography system being used by the device [117]. b) Open Debugging ports: The potential attackers can easily exploit vulnerable network services running on the target device through open ports. The manufacturers ship most of the IoT devices without disabling their debug ports. These ports could be used to take full control over the system. The intruder could inject malicious code, modify the firmware, bypass the security, spy, and bag their data. Hence, a plethora of attacks could be launched through these open ports [118], [119]. For instance, most of the botnets like Mirai, BrickerBot exploit telnet port. BrickerBot, another IoT botnet unveiled by researchers at Radware in April 2017. This malware launched a permanent denial-of-service attack, which prevents the prey's hardware from functioning. To create a botnet, it included all the devices exploited by Mirai or other botnets. These devices were with open SSH port (22) and older versions of SSH Server. It also targeted the devices with open Telnet port (Linux /busybox based). Furthermore, it leveraged the default login credentials by consistently attempting 'root'/'vizxv. It can affect in many ways, for example, can corrupt storage [14]. c) No energy Harvesting: The sparse resource nature of IoT nodes makes them vulnerable to resource exhaustion attacks. Moreover, there is no mechanism to harvest the energy of these low-power IoT nodes [120]. These attacks could jam the communication channels and can cause extensive unauthorized utilization of IoT resources like bandwidth, memory, CPU time, disk space. It leads to battery drainage of IoT nodes, and they could not provide their services to legitimate users [10]. By exploiting this vulnerability, the adversary can launch the battery drainage attacks, sleep deprivation attacks, Node outage, DoS attack, etc.
• Battery drainage attacks: IoT nodes work with low-power battery and that too with no recharging mechanism. The intruder floods the node with so many legal requests that it ends up with exhaustion. The number of attacks can be launched by draining the energy of a node [122].
• Node outage: It prevents the edge nodes from performing their function in the system. The number of factors like-battery drainage, code injection, unauthorized access, sleep deprivation could lead to this attack [110]. d) Weak Authentication Mechanisms: Implementing strong authentication mechanisms at different interfaces like mobile, cloud, and web in the IoT ecosystem makes them more secure. The adversary targets these insecure interfaces through weak credentials and account enumeration. If anyone can access the IoT nodes without undergoing identity checks or bypassing the weak authentication system, then the adversary can exploit the system in numerous ways [123]. The adversary may launch DoS attack, steal data, and take complete control of the system. It becomes difficult to implement strong authentication mechanisms due to a lack of adequate resources for IoT devices. Under such conditions, the authentication keys must be exchanged and stored securely to ensure effective authentication. The number of attacks could be framed out by exploiting this vulnerability.    [14].
• Dictionary Attack: It is a brute force attack in which the intruder enters/access the IoT system/device by trying all the combination of characters in the dictionary to break the security.
Kolias et al. [14] described how the dictionary attack leads to highly distributed attacks. It creates an army of millions of infected nodes by compromising their security to launch attacks like DoS, DDoS.
• Sybil Attack: It causes the malicious node to impersonate the real IoT node by manipulating the identity of compromised nodes. This attack adversely affects network performance. The forged device may flood channels with false packets to make the services of IoT system inaccessible to authorized users, can eavesdrop on traffic, fabricate the message and much more [124].
• Hello flood and Homing Attacks: In this attack, the adversary leverages the fact that the new node sends the ''HELLO PACKETS'' among all the neighbors when it is added to the network to indicate its arrival. At this, all the receiving nodes assume the sender node to be in their communication range. For this attack, the attacker uses the node with higher transmission power [125]. e) Improper Encryption: The integrity of smart applications depends upon the security of data collected by widespread sensors. The data must be safe and verifiable both at source and in transit.Most of the IoT devices use less reliable wireless communication media, for example, NB-IoT, Zigbee, SigFox, LoRa, 802.11. a, and 802.15.4. As a consequence, these devices are more susceptible to data leakage attacks [126]. It is also found that most of the IoT devices collect the personal information of users to provide essential services. In the case of e-health, misuse of this personal information can be life-threatening and, in some cases, unavailability of scathing health services [127]. Encryption is an efficient way to protect the users' data from disclosure to unauthorized users. However, the resource limitations of IoT devices makes it difficult to develop a robust crypto-algorithm. As a result, the adversary can easily dodge the deployed crypto-algorithm. This may lead to many more serious attacks [10].
f) Unauthorized Access: To prevent unauthorized access to an IoT system, a secure credential management system should be implemented. Today the market is flooded with IoT devices having hardcoded, weak, or default credentials. No one takes care to change the default credentials and there it creates a hole for attackers. In the most significant DDoS attack, namely Mirai, the attackers used around 60 default login credentials to turn IoT nodes as the malicious one. It reached to 1.2 TB per second [14]. Furthermore, baby monitors were shipped with hardcoded credentials, which could only be fixed by firmware patch [17]. g) Insufficient Audit Mechanism: There is as such no mechanism to maintain the logs of what has been done in the IoT devices and check them time to ensure their security [128]. The events like application errors, successful/failed login attempts, authentication attempts, authorization attempts should be logged in an encrypted log file. h) Improper patch management: The operating system of IoT devices and their firmware must be updated regularly so as to augment its function and protect against attack vectors. Albeit it is found that in IoT devices, the manufacturers skip the automated patch-update feature. Moreover, they do not provide the security patches on time and that too may lack integrity assurance. The insecure updates may contain malicious patches which can adversely affect the devices at large [129]. To ensure the security of updates,the update file must be verified, signed, encrypted, and transmitted via a secure connection. i) Boot process Vulnerabilities: During the boot process, all the three -firmware, boot loader, and boot process sequence is vulnerable to get leverage with. For instance-In an experimental setting, the researchers launched such attacks against the Nike+ fitness tracker and Nest Thermostat [130].
Google Nest Thermostat got compromised over this loophole, where the adversary loaded the thermostat with a malicious initial boot loader along with a custom full boot loader and an argument list for the onboard kernel. The arbitrary payloads like backdoor could be added later on using a custom loader. They made the processor to boot from UART or USB interface and inserted their boot loaders. They even made it possible to accept updates from a source other than Nest [128].

IV. TAXONOMY OF IoT VULNERABILITY ASSESSMENT TECHNIQUES
The unique traits of IoT devices with its growing number have made it difficult to continuously figure-out evolving IoT-specific vulnerabilities. Furthermore, attackers are getting more skilled in launching stealthy attacks. To secure IoT and to make it more resilient, the security mechanisms must include the regular vulnerability assessments as an integral part. In this context, we have explored various monitoring and security assessment strategies, which are depicted by FIGURE 10 and described as follows. a) Security Testbeds: The security evaluation of a system before its deployment helps in discovering IoT vulnerabilities before their exploitation. were designed to identify vulnerabilities in different set-ups. The previous testbed works in a laboratory with physical devices, and the latter uses python scripts to emulate device behavior. The authors claim that both the virtual testbeds are inter-operable with real control systems and efficiently emulate real systems in terms of threats too. In another work, the authors [133] presented a scalable framework, named as Smallworld. In this platform, several scenarios are made using simulation and virtual environments to find the vulnerabilities within the IoT system. The proposed platform consists of five layers, namely, perception, abstraction, services, API, and management layers. Further, the authors considered a case study on home automation applications using virtual and real smart devices to show the effectiveness of the proposed work. Additionally, Siboni et al. [134] proposed the IoT testbed integrated with multiple plugins for penetration testing. The architectural model of testbed works in four modules; a module for management and control actions, a module responsible for actual testing sequence, a module to execute a set of security tests, and the last for measurements and analysis. The Security Testing Module provides the number of plugins for security tests like Fingerprinting, Port Scanning, Process Enumeration, Communication Tampering, and Vulnerability Scan. The IoT devices are tested in multiple test scenarios in the proposed testbed.
As an extension of this work, the authors plan to implement the proposed security testbed with testing systems, such as an IoT-based honeypot environment. On similar lines, Siboni et al. [135] proposed a testbed specifically on wearable devices. The vulnerability assessment is also one of the modules under various IoT testbeds [136]. TABLE 8 provides an overview of the IoT testbeds. b) Machine Learning: Machine learning is a part of an umbrella term AI that provides the machine with the ability to learn from experience, examples, and analogies [140]. As learning occurs, the machine becomes more intelligent and capable of making informed decisions. The objective of machine learning is to efficiently imitate human learning activities by computers such that the knowledge can be automatically discovered and acquired. Several machine learning algorithms have been widely applied to improve IoT Security. A learning algorithm is one that takes as an input a training set and tested upon the testing dataset. Some of the widely used machine learning tools for implementing learning algorithms are described in TABLE 9. In general, there are three main categories of learning: supervised, unsupervised, and reinforcement learning [141], as shown in FIGURE 11. FIGURE 11. Machine learning techniques [140], [141].
The resource-constrained IoT devices can neither rely on existing security solutions nor lightweight security mechanisms. The former class suppresses the IoT nodes with computation and communication freightage. The latter opens the doors for intruders to enter with ease. Machine Learning is a promising solution at this end. Moreover, the main element in Machine Learning is data. The widespread IoT led to the generation of enormous data regularly, which can be concluded as a goldmine for machine learning. The intelligent system learns from a massive amount of data and provides high efficiency and considerable accuracy with minimum computation cost. In [47], the authors stressed the usefulness of ML in IoT in terms of its scope, security, and inferring insights from data. The significant applications of ML-like detecting outliers, pattern recognition, feature extraction, and predicted values are essentials of IoT security. The review of nascent solutions based on machine learning to cope with growing attacks are discussed in the following sub-sections and shown in FIGURE 12. Azmoodeh et al. [148] proposed an OpCodes based deep learning method to detect malware on the Internet of Battlefield things. The authors used a class-wise information gain method for feature selection (OpCodes sequence). The VirusTo-tal3 Threat Intelligence platform was used to collect the malware samples between February 2015 and January 2017 [149]. Furthermore, the official IoT App stores (Pi Store 4) were used to collect goodware samples. The opcodes were extracted by using objdump. It is revealed from their research that the proposed approach with accuracy rate (AR) 99.68 percent, precision rate (PR) 98.59 percent, and recall rate (RR) 98.37 percent outperforms the other approaches [150], [151].
Hasan et al. [152] proposed the Multiclass Classification model for attack and anomaly detection using several ML Approaches.    [165], [166]. The authors have also shown that the memory requirement of RNN models can be reduced by a factor of 5 by using linear quantization-based compression technique without compromising accuracy. Moreover, the performance can be improved using GPU offloading approaches. Jayasinghe et al. [167] proposed a quantifiable trust accessibility model. As per the authors, trust plays a crucial role in the successful future of IoT. The authors have used a numerical method to extract basic trust features. Furthermore, they applied unsupervised machine learning algorithms: k-means clustering to label the interactions as trustworthy or untrustworthy based on the trust mentioned above attributes. They calculated the optimal cluster size using Elbow Method and further used Principal Component Analysis to reduce the size of the training matrix. To bring the data samples in the range of 0 and 1, features were normalized. They used Radial Basis Function Kernel to get the non-linear boundary which can separate the trustworthy and untrustworthy interactions and learn how to combine trust Attributes in the best way to obtain a final trust value.
In the IoT based Grid system, machine learning techniques are applied for proper analysis of the large volumes of data and thus aids in decision making to run the grid (smart). These techniques proved to be useful in a number of ways, such as the prediction of power consumption, price, power generation, detection of network intruders, future optimum schedule [151], [168], etc. Li et al. [169] analyzed user proclivity in a smart grid by applying machine learning techniques to find usage patterns. Remani et al. [170] applied reinforcement learning for scheduling residential load taking into cogitation renewable energy sources as well. For short term prediction in terms of load forecasting in smart grids, deep neural networks are used by Ryu et al. [171], [172]. TABLE 10 presents the summary of ML-based solutions to assess the vulnerabilities and to secure the IoT system. Moreover, Machine learning has been increasingly used in predicting vulnerabilities like Cross-Site Scripting, SQL Injection, file inclusion vulnerabilities, remote code execution in web application [139], [175]. Even some of the proposed platforms also provide the feature of vulnerability correction [139]. There is an ongoing project named the High Assurance Cyber Military Systems program, announced by DARPA in 2012 in the US to patch the vulnerabilities of IoT, particularly the military vehicles, drones, and medical equipment [35]. c) Fuzzy Techniques: A research direction in this realm applies fuzzy-based approaches to assess the security of IoT protocols. Lahmadi et al. [176] designs one such framework, which evaluates the 6LoWPAN protocol. The proposed testing suite employs the mutation algorithms on messages on the network layer, to find the deviation of actual responses of IoT nodes from the conventional ones. On similar lines, in [177], a fuzzy technique is applied on Zigbee networks to find the vulnerabilities within the network. The proposed technique is the combination of Finite State Machine with Structure-based fuzzy algorithms. After conducting a series of performance tests, the authors claim that the proposed finite state machine-based algorithm is more accurate than a structured-based algorithm. In another noticeable work, the authors have given a fuzzing tool EUFuzzer which discovers the vulnerabilities in human machine interfaces [178]. Several graph -based solutions have also been given to find the vulnerabilities and secure IoT networks [179], [180]. d) Honeypots: The honeypots trap the adversaries by imitating real IoT assets but having no value for them, by calculatedly creating security vulnerabilities.
With the help of honeypots, we can determine the strategies and attack paths used by attackers to carry out malicious activities [181]. In context to IoT, honeypots, generally imitates a specific type of IoT device to further scrutinization of attack vectors in a particular environment. The IoT honeypot, namely HIoTPOT [182], finds that most of the attackers are interested in finding vulnerable devices, as per the analysis of per-day traffic. One of such honeypots, IoTPOT [183], mimics Telnet services of several IoT devices to investigate the current attacks profoundly. The authors observed that the Telnet-based attacks are carried out in three phases: intrusion, contamination, and monetization. In the first phase, numerous login attempts are made by an attacker with combinations of credentials. Following this in the next phases, malware is downloaded in the device which is spread across the network to launch a DDoS attack. The authors also tracked the several malware binary files downloaded and proposed IoTBOX, for analyzing captured malware binaries. Another honeypot, HosTaGe, is designed by Vasilomanolakis et al. Authors of [184] target malicious activities against several protocols like Telnet, HTTP, SSH, FTP, MySQL, and SIP in ICS networks. HosTaGe also generates the attack signatures, which can be integrated into ICS for further detection and mitigation of attacks. On similar lines, the authors proposed a honeypot, Cryplh [185], to find the attacks against the PLC-based ICS.
Litchfield et al. [186] designed a HoneyPhy honeypot, hybrid-interaction based CPS framework, which can imitate the behavior of both IoT devices and CPS processes. The proposed honeypot comprises of three modules. A module to maintain connections and traffic, the process module to imitate the systems' dynamics, and device models to mimic CPS devices. Similarly, Guarnizo et al. [187] proposed the IoT-based honeypot framework, SIPHON, which attracts malicious traffic on the internet through wormholes and vulnerable IoT devices. The authors concluded with insightful inferences regarding malicious traffic, ports, target location, and user agents. Additionally, honeypots [188] VOLUME 8, 2020 were designed to analyze attacks against a ZigBee gateway. In [189], the authors reported most of the attacks against the Zigbee-based IoT devices were the dictionary attacks. e) Penetration Testing and Network Discovery Tools: Visoottiviseth et al. [190] developed a penetration testing tool PENTOS for IoT devices. It compiles various penetration tools like Metasploit, Kali, Nessus, Burpsuit, Cain & Abel, etc., to find the vulnerabilities. It also guides the users against OWASP's IoT vulnerabilities. There are many features in PENTOS, which aids in gathering information, scanning the web, Bluetooth analysis, and reporting. The authors paving in this way as Chen et al. [191] suggested the path of intelligence and modularization for penetration testing tools to discover vulnerabilities by employing offensive attacks against IoT systems. The authors carried out PT in three modules. The interface testing checks the interfaces through which multiple devices or users interact with. The transportation testing targets weak cryptographic schemes, misuse issues, and flaws in communication protocols. The system testing focuses on firmware, OS, insecure system settings, etc.
To scan the whole IoT space, Markowsky et al. [192] three scans: Shodan to target the Cayman DSL Routers, Masscan to target the devices affected with Heart bleed bug, and last they used Nmap with PFT to target vulnerable connected printers. The authors [193] also proposed a management platform where, after the vulnerability assessment, the information about the same is shared with the users. They start with collecting the device information and then comparing it with vulnerability information from the National Vulnerability Database. The information is collected through IP Scan, Handshake Scan, and finally, OS Fingerprinting. For sharing the information about vulnerabilities Structured Threat Information Expression standard is used.
We have seen that new vulnerabilities and attacks are evolving every day [194]. It is found that there are a few vulnerability assessment solutions based on Machine Learning. Recently Dojo by Bullguard introduced BullGuard launches intelligent IoT vulnerability scanner [195]. The dojo is available for both android and ios. It scans all the IoT nodes in the wi-fi network, analyzes the vulnerabilities and scores them according to the risks they are undergoing. This is the first machine learning-based tool. There are some other tools, for instance, Bitdefender, IotSploit, IoTScanner, Shodan [196] Censys, SeeSec -IoT Vulnerability Scanner that serve the same purpose.

V. CASE STUDY: SMART AGRICULTURE
In order to highlight the need of sustainability in Smart agriculture, we consider a case study on ''Sustainable Smart Agriculture''. The contextual analysis of the same is covered.

A. SUSTAINABLE SMART AGRICULTURE
The advanced technologies like IoT, robotics, cloud computing, artificial intelligence, unmanned aerial vehicle, and machine learning have replaced the conventional methods of farming with modern methods to maintain the supply-demand ratio. The proposed use case (Sustainable Smart Agriculture) sustainable framework is shown in FIGURE 13. The way it works could be well-understood in context to layers. The physical layer generally covers the field with several underground and above the surface sensors, drones, tractors, pesticides, and fertilizer controllers. The on-field devices communicate with each other and the local control center at the edge layer through gateways. These devices gather real-time information regarding soil, weather, livestock, energy management, and irrigation. From the edge layer, the information collected by sensors is sent to the cloud for further analysis. The useful insights inferred are sent back to the user/owner for further actions. The edge nodes provide various services like real-time monitoring, security mechanisms, energy harvesting, and prediction at the edge layer [197]. For example, ML-based models classify the events related to plants or livestock, predict the crop yield, fertilizer needed, and schedule the irrigation based on water needed by the crops. And the network layer facilitates communication among all the layers [198]. The smart services with advanced technologies provide manifold benefits to the agriculture industry but expose the risk of vulnerabilities and cyber-attacks too. For example, the surveillance drones in the field capture images of the crops, and with computer vision, the disease-prone area of the field is found. The identified affected area is then isolated and removed to further prevent the entire field from the disease. This sector being unaware about the cyber-security is more prone to cyber-threats. The intruder can remotely take control of on-field sensors, access irrigation management system, and maliciously manipulate the data in transit. With such attempts, they can destroy the field of grown crops, control the drones to damage the crops through pesticides or over fertilizers, and create an unproductive agricultural environment. Thus, less productivity even deteriorates the economy of a country [199]. The cyber-attacks targeting smart agriculture when launched in a well-coordinated manner on a large scale called agro-terrorism [200]. The data and malware injection attacks are the most prevalent attacks in smart farming. In a data injection attack, an intruder maliciously modifies the data contributing to real-time decision making and thus leads to false decisions. In malware injection attack, an adversary infects the smart device by injecting a malware, that has self-propagating like features. Consequently, malware is most likely to infect all other smart farms having similar deployments. It can turn the devices into bots, steal information, and hamper the functioning of smart machines in the field [14].
A feasible solution to overcome the aforementioned sustainability issues is the incorporation of energy harvesting techniques and vulnerability assessment in smart agriculture. The vulnerability assessment module added to the Smart Agriculture has four components namely collector, analyzer, quantifier, and mitigator. The collector collects all the information related to the device and network. The analyzer takes the collected information as input and compares the same with NVD. The multiple vulnerability assessment techniques like machine learning, network discovery tools, penetration testing tools, testbeds, machine learning models, and honeypots can also be a part of this component. The quantifier quantifies the vulnerabilities assessed by analyzer using CVSSv3 as per the risk they impose on the IoT system. The mitigator gets the vulnerabilities with their CVSS score as input and provides solutions to overcome those loopholes. It also notifies about the vulnerabilities and their remediations to the manufacturers and IoT users. The timely mitigation of the known vulnerabilities reduce the risk of the IoT system as prey to potential adversaries.
For the sustainable functioning of IoT nodes in smart farms, both the underground and above the surface sensors need the continuous power supply for sensing and communication. The ideal lifetime of sensors in an agricultural field is found to be more than 5 years. Though the power requirements have been reduced in the sensors, with the recent advancements in technology and sensor materials. But, they need continuous powering to enable communication between the underground radios and above-ground receivers. There are many EH sources like underground living plants, vibration, thermal, solar, and bacteria by way of fuel cells to harvest the energy from the surrounding sources in the field. The other way to power these energy-constrained sensors is through wireless power transfer, which can transfer both the data and power in a full-duplex mode. For example, the above the surface nodes transfers the harvested solar energy to the underground nodes, and the other way could be the underground nodes transfers the harvested energy from bacteria and vibration to above the surface nodes. Thus, this bi-directional power transfer results in longer battery life for sustainable smart farming [201]. However, the research community needs to work on wireless RF-based underground power transfer for sustainable IoT. Additionally, the factors like maintenance, battery replacement, repair, and underground re-installation must also be worked upon. The development of smart farm technologies, therefore demands further research before wide adoption in the community. Integrating both the components in smart agriculture has several benefits: • Reduce the risk of attacks and make the system more secure with a prior vulnerability assessment.
• Improve the lifetime of power-constrained sensor nodes with EH schemes integrated at the design phase.
• Prevents the privacy of data at storage and in transit with light-weight cryptographic schemes.
• Protect the system from misleading data injection attacks.
• Reduces the smart nodes from being compromised.
• Aware the developers with the vulnerabilities in the shipped devices.

VI. OPEN ISSUES AND RESEARCH CHALLENGES
So far, we summarized the several IoT attacks launched by exploiting common vulnerabilities in an IoT system, along with few vulnerability assessment mechanisms and ML-based solutions to secure IoT systems. Further, we discuss the emerging challenges for sustainable IoT as shown in FIGURE 14 and pinpoint some initiatives for future work, to be pursued in this vital field of IoT sustenance.

A. LACK OF SCALABLE VULNERABILITY ASSESSMENT WAY OUT IN IoT SYSTEM
The key technologies in IoT, for instance, RFID [49], WSN are themselves vulnerable to threats like node compromise, eavesdropping, tracking of devices, malware, etc. Moreover, numerous attacks are exploiting the unanticipated vulnerabilities in IoT systems. Although there are several vulnerability assessments, they are not mature enough to deal with the heterogeneous network of IoT devices. There must be an automated vulnerability assessment framework to handle device scalability and variability in various deployment contexts. Alongside, there must be the mechanism to find the unexploited vulnerable device. Future Initiative: Transfer learning algorithms context [202] could be a potential solution at this end. Feeding the TL algorithms with IoT vulnerabilities could enhance and automate the job of vulnerability assessment so as to infer this knowledge from numerous IoT devices. Furthermore, investigating IoT-specific trust models [203] in several contexts would aid the growth of requisite IoT remediation strategies after assessing the vulnerabilities.

B. UNEXPECTED USAGE OF IoT DATA
The IoT has enabled the ubiquitous computing, and thus deployment of IoT enabled sensors in our lives have been noticeably pervasive. These sensors collect a huge amount of data from its surroundings and transmit it to the cloud for further processing. The value of the IoT system lies in that data [204]. Although, the privacy of the collected data is of main concern, static as well as in transit. For instance, IoT based baby monitors and IoT toys were easy to play with by hackers to get sensitive information like video streaming of baby monitors [17], voice recordings of parents, and their kids(in millions), emails, passwords, etc. Recently, it is unveiled that a lot of privacy-sensitive information could be revealed from varied types of related data (user/environment) gathered by smart sensors. Thus potential adversaries could make useful insights from the collected data in an unexpected manner. For instance, privacy-sensitive information like daily routines, the number of persons in a home can be deduced by analyzing smart homes' power usage data collected using smart meters [205]. Some serious consequences that result by providing 3rd party applications access to sensors are discussed in [206]. Despite such previous attempts, the amount of sensitive information that might be deduced from probably shallow data is not well-known or well-understood yet.
Future Initiative: The privacy-friendly techniques must be incorporated into Smart Meter Architecture for making insights out of the sensitive collected data. The authors [205] suggested pertinent data selection or processing methods to minimize or avoid sensitive personal data within industrial applications.

C. LACK OF CAPABILITIES TO AWARE IoT USERS ABOUT SECURITY GUIDELINES
Further, there is a need to explore the ways to improve users' awareness about the serious consequences of numerous IoT threats. A noticeable example is a DDoS attack using Mirai malware launched in October 2016, targeted against the service provider, Dyn and shut down the hundreds of websites for several hours-Netflix, Twitter, GitHub, Reddit. Mirai launched the DDoS attack against target servers by building a network of weakly configured IoT nodes, namely botnet [207]. While the code of Mirai malware was accessible months ago, and despite some vendor's reaction, most of the end-users did not bother to update their IoT devices with security patches, thus permitting them to be part of many upcoming botnets [208]. Note that although the vendors produce the security patches, the updates need not reach the requisite destination. Even today, the same vulnerabilities of IoT devices are exploited by Mirai and its variants [14], resulting in severe consequences like full device control.
Future Initiative: There must be some mechanism to constrain the users to modify their default credentials and automate the process of frequent firmware updates. Additionally, there must be some limit on the time gap from the moment the vulnerability discovered in producing patches and then deploying to the IoT attack victims for sustainable IoT.

D. LACK OF ENFORCEMENT OF MINIMUM IoT SECURITY STANDARDS
The need for standardization processes for different IoT products and various IoT applications question the security, interoperability, scalability, and compatibility in the IoT realm. The manufacturers supply IoT products without considering baseline IoT security standards [39], [209]. They mainly focus on their functionality, low cost, and low energy consumption. Whereas, the current status of IoT threats and growing attacks, emphasizes the incorporation of various security schemes in IoT devices. These schemes include hardware security against tampering, indispensable user authentication, data encryption at rest and transit, and OS/firmware/application security and integrity. However, the resource constraint nature of IoT devices like sensors, baby monitors, CCTV cameras, and the computation and memory overhead of traditional security schemes, conveys the necessity of lightweight security solutions for IoT devices [210].
Future Initiative: There must be an international body for enforcing minimum security standards in heterogeneous IoT products and IoT applications. Additionally, it is recommended to have transparency, communication, and collaboration among evolving IoT security standards-making bodies like IoTSF, Industrial Internet Consortium, NIST, and International Society of Automation [50].

E. ML-BASED VULNERABILITY ASSESSMENT CHALLENGES
Machine learning has shown its significance for both, one way it extracts the value from the data, and the other way used by the adversaries for malicious purposes. The features of these learning algorithms are being misused by potential adversaries to breakdown the cryptographic schemes. For instance, previous studies [211], [212] used SVM to collapse the cryptographic systems. Another study [213] concluded that RNNs could do cryptanalysis by capturing and learning the algorithmic demonstration of polyalphabetic ciphers. Additionally, researches have shown that feeding false data can compromise the training models and hence, failing the entire system. For example, if deep learning-based model for controlling the self-driving vehicles is injected with false data, the malicious user can potentially control the vehicle [54]. Moreover, there is the risk of adopting poor defense policies during the initial stage of the learning process, which can lead to security disasters in learning-based IoT security system. Additionally, supervised and unsupervised learning now and then fail to identify the malicious activities because of oversampling, not enough training data, and poor feature extraction. Hence, the ML techniques need to be integrated with back-up security mechanisms. Moreover, several existing ML techniques have exhaustive overheads, so new security schemes such as dFW must be explored for building secure IoT systems, particularly for the situations when there exists no cloud or edge computing.
Future Initiative: Applying transfer learning [214] to actual defense experiences reduces accidental exploration, increases the learning rate, and reduces the chances of having poor defense policies in the initial stage of learning. Also, ensemble ML techniques [47] prove to be more fruitful in overcoming loopholes of basic learning schemes.

VII. CONCLUSION
IoT has evolved with immense growth in participating entities, i.e., sensors, communication, and computation. IoT is in its way to transform all the major aspects of our lives from homes to health, to agriculture, to automation, to cities, to transportation, to grids and manufacturing.
This revolutionary expansion will be useless if IoT is not able to sustain in the present situation, with a void in energy-efficiency and security in the present ecosystem design. In this article, we have provided insights to the readers about the Sustainable IoT, embedded vulnerabilities in IoT devices, and vulnerability assessment techniques to assess those vulnerabilities before getting exploited. This article is divided into four parts. The first part discussed the general concepts related to IoT. In this, we begin with the background of IoT and explored many significant events related to its growth since this term emanated. We discussed the factors for the sustainability of IoT, the protocol suite, and testbeds. The second part discussed the IoT security vulnerabilities such as open ports, poor update mechanisms, and weak authentication practices serving as entry points for attackers causing malicious abuse. Then, we explore the contribution of machine learning, security testbeds, honeypots, and network discovery tools in assessing vulnerabilities in an IoT environment. We have also presented the case study on sustainable smart agriculture. Then, we listed the open issues and future initiatives for sustainable IoT. Finally, we summarize that this article provides useful insights to the research community by presenting the present-day status of such a vibrant area of research.
In future, hybrid ML techniques and deep learning will be explored in detail for Vulnerability Assessment in IoT. The attacks on EH chips and their consequences will also be studied. We will also try to cover the frameworks for quantifying the vulnerabilities in IoT and its real-time implementation in multiple scenarios.
POOJA ANAND is currently a full-time Research Scholar with the Department of Computer Sciences and Information Technology, Central University of Jammu, India, under the supervision of Dr. Yashwant Singh and Dr. Arvind Selwal. Her current research interests are in the areas of operational cybersecurity, including attack detection and characterization, vulnerabilities assessment methodologies, the Internet of Things, and machine learning.
YASHWANT SINGH received the bachelor's degree from SLIET Longowal, the master's degree from Punjab Engineering College, Chandigarh, and the Ph.D. degree from Himachal Pradesh University, Shimla. He was with the Jaypee University of Information Technology for 10 Years. He is currently an Associate Professor and the Head of the Department of Computer Science and Information Technology, Central University of Jammu, where he has been a Faculty Member since 2017. He is also a Visiting Professor at Jan Wyzykowski University, Polkowice, Poland. He has collaborated actively with researchers in several other disciplines of computer science, particularly machine learning and electrical engineering. Roughly, he has served on 30 conference and workshop program committees and served as the General Chair for PDGC-2014, ICRIC-2018, ICRIC-2019, and ICRIC-2020. He currently serves as the Coordinator of the Kalam Centre for Science and Technology (KCST), Central University of Jammu, established by DRDO. He has published more than 70 research articles in international journals, international conferences, and book chapters of repute. He has research projects worth INR 15 Crore (Approx.) in his credit. He has guided three Ph.D. and 24 M.Tech. students. He has visited eight countries for his academic visits, i.e., U.K., Germany, Poland, Chez Republic, Hungary, Slovakia, Austria, and Romania. His research interests lie in the area of the Internet of Things, wireless sensor networks, ICS/SCADA cybersecurity, ranging from theory to design to implementation.
ARVIND SELWAL is currently working as the Sr. Assistant Professor with the Department of Computer Science and Information Technology, Central University of Jammu, India. He has more than 14 years experience of teaching UG and PG classes. He has successfully guided 12 M.Tech. and one M.Phil. students. His research interests include machine learning, biometric security, image processing, and advanced database systems. He has contributed more than 20 research articles in reputed international/national journals indexed in Scopus and SCI. He has authored a book on Fundamentals of Automata Theory and Computation.
MAMOUN ALAZAB (Senior Member, IEEE) received the Ph.D. degree in computer science from the School of Science, Information Technology and Engineering, Federation University of Australia. He is a Cyber Security Researcher and Practitioner with industry and academic experience. He is currently an Associate Professor with the College of Engineering, IT and Environment, Charles Darwin University, Australia. His research is multidisciplinary that focuses on cybersecurity and digital forensics of computer systems with a focus on cybercrime detection and prevention. He has more than 150 research papers in many international journals and conferences. He is the Founding Chair of the IEEE Northern Territory (NT) Subsection.  IEEE TIE, IEEE TDSC, IEEE TITS, IEEE TCC, IEEE TKDE,  IEEE TVT, IEEE TCE He is leading the research group Sustainable Practices for the Internet of Energy and Security (SPINES), where group members are working on the latest cutting edge technologies. He is a TPC member and reviewer of many international conferences across the globe. VOLUME 8, 2020