Investigation of Machine Learning-Inspired Security Controls Used in Protecting Health Critical Infrastructure from Advanced Persistent Threat Attacks: Review

The cybersecurity of National Critical Health Infrastructure is of utmost importance. If its security is not prioritized, it can directly impact the safety and well-being of patients, healthcare professionals, the general public, and the economy of the affected state. National Critical Health Care Infrastructure is a complex system involving multiple components essential for providing quality healthcare services to individuals in a country. To enable seamless access to healthcare services, EMRS have been implemented across the globe and thus cyber physical systems have become the medium of service delivery. With the criticality of these systems in mind, should a cyber-attack occur, all parties dependent on these infrastructures will be negatively affected. There are myriad cyber-attacks, such as advanced persistent threat attacks targeted at critical national infrastructure, which adversely compromise the security of the technologies in such infrastructures. The critical healthcare infrastructure is no different from other cyber physical critical infrastructures and thus has been a target of Advanced Persistent Threat (APT) groups for an extended period now. Machine learning solutions have been used to effectively protect, defend and respond to cyberattacks in software defined networks, SCADA for instance. As a result, this paper intends to ascertain the machine learning-inspired security controls that can be used to protect National Health Care Infrastructure against APT attacks. This review follows a mixed-method systematic literature review to answer the research question. The study's results reveal that machine learning has been majorly employed in detecting APT in critical infrastructure, not health-critical infrastructure and that the resilience of machine learning-inspired security controls has not been thoroughly researched for cyber health care systems. Even though machine learning has been tremendously applied in identifying APT, numerous challenges still exist, such as the data heterogeneity and complex strategies used by the APT groups, making it difficult for conventional security controls to mitigate APTs fully.