Abnormal Network Traffic Detection using Deep Learning Models in IoT environment

As an emergent technology, the internet of things (IoT) aims to create an interconnected world of smart devices autonomously communicating via the internet. Integrating heterogeneous devices and networks increases the number of threats and attacks in such environments. Thus, it becomes necessary to detect any network anomalies that may be a sign of such attacks. Traditional classification algorithms are not very competent at solving network traffic anomalies due to massive data. Deep learning techniques have demonstrated their efficiency in this aspect by performing accurate detection due to their capability of extracting and learning better features from the data and classify unknown attacks. In this paper, a two-stage data assessment for anomaly detection of IoT network traffic is proposed. The first stage is data analysis. In this stage, the data is filtered and classified where the main features are selected for the train and test process. The second stage is anomaly detection; where two well-known neural network algorithms are deployed namely; Long short-term memory (LSTM) and Feedforward Deep Neural Networks (FDNN) algorithms. A real dataset is used to evaluate the efficiency of the used algorithms where a set of parameters are evaluated. The main findings of this work are that the investigated models are suitable for binary classification and can achieve high detection accuracy with 90.66% for LSTM and 64.12% for DFNN.