Detection of Blackhole and Wormhole Attacks in Wireless Sensor Networks Combined and Compare with Machine Learning and Deep Q-Networks Reinforcement Learning

Wireless Sensor Networks (WSNs) are exposed to several security threats; blackhole and wormhole attacks compromise the correct functioning of the network and the integrity of the transmitted information. The goal of this work is to identify these attacks using a hybrid of supervised and deep reinforcement learning algorithms. MobileNetV2 with KNN, Random Forest and Decision Tree algorithms were used together with Deep Q learning to improve detection accuracy and reliability. The proposed models were benchmarked on detection rate, classification rate (ability to discern between blackhole and wormhole attacks), false positives/negatives, computation costs, and expansibility. It was established that DQN offered improved detection accuracy and classification more so than other traditional machine learning models although the false positives and false negatives were negligible. The results of the presented simulation show the ability of the hybrid approach to scale well for larger network sizes at reasonable rates of computation. Thus, the presented results reveal the possibility of further developing the approach based on the integration of machine learning and reinforcement learning to control the presence of threats in WSNs and achieve more secure network infrastructures.