Hybrid Feature Selection and Classification using RF-DNN for Anomaly Detection in IoT-WSN

Abstract

There are a number of factors that may impact how an attack detection system can identify a threat. It is clear that current Intrusion Detection System (IDS) approaches for the Internet of Things (IoT) are still in their youth. There are just a few ways to categorise attack types.  However, only conventional networks have applied and assessed such techniques. Due to this, the IoT-specific needs and computational capabilities of these approaches were not taken into account while developing these methods. In this paper, hybrid feature selection and classification using Random Forest-Deep Neural Networks (RF-DNN) for anomaly detection technique in Internet of Things (IoT) Wireless Sensor Networks (WSN) is proposed. In this technique, a filtering method of Fisher’ score and correlation coefficient is applied to select the candidate feature set.  Then the combination of RF and DNN is used as the classifier for feature selection.  The static properties are divided into five primary categories. Similarly, it categorise the dynamic features into the classes of location, network, protocol, registry and Internet Protocol (IP) address. Experimental results show that the proposed RF-DNN algorithm achieves higher detection accuracy, higher throughput, lesser computational cost and  higher residual energy, when compared to the  existing techniques.