Enhancing Intrusion Detection in Wireless Sensor Networks through Machine Learning Techniques and Context Awareness Integration

Abstract

Wireless sensor networks (WSNs) play an imperative role in the communication among independently implemented wired, localized, or mobile sensors in a cyber-physical system. Environmental monitoring, object identification, data acquisition, analysis, and transmission to the owner of the wireless sensor network are the primary research focuses in this work. Due to the flexibility of WSNs and the scarcity of resources in IoTs, both are frequently integrated. But the above integration makes these networks more open to outside factors and, thus, vulnerable to attacks like flooding, assaults, grayhole attacks, black hole attacks, and other scheduling problems that are typical in such circumstances. There are certain inherent properties of Wireless Sensor Networks (WSNs) that have made the detection of intrusion unsuitable, namely; false alarms, it has very high computing overhead and a poor detection ratio. This problem has arisen due to the network’s excess of data tightly connected in a dense manner and resource constraints of sensor nodes. For that purpose, this research recommends the utilization of machine learning techniques for intrusion detection in WSNs. The detection precision is enhanced by enhancing the use of Support Vector Machine (SVM ) combined with stochastic gradient descent (SGD). Moreover, the research proposes the integration of context knowledge, which is known as context awareness that takes into account user preferences and system characteristics or situations to improve the performance of the recommendation systems. In order to decentralize the computational load of the system, the first traffic data is reduced by principal component analysis (PCA) and singular value decomposition (SVD). The network risks observed are further categorised using an VG-IDS model. Worthy of note is the fact that the recommended WSN-DS algorithm was capable of yielding better results than other complex algorithms that were examined by using the WSN-DS dataset throughout the evaluation process; the accuracy rate of the proposed WSN-DS algorithm was 96%. It is seen in improvement of accuracy, and recall and F1-measure rates to the improved figure of 98%, 96%, and 97% respectively.