Performance Analysis of Machine Learning Algorithms in Detecting and Mitigating Black and Gray Hole Attacks

Abstract

Listen

Translate article

AbstractBlack hole attacks and gray hole attacks are major denial of service (DoS) attacks in wireless sensor networks (WSNs). Intrusion detection systems are proposed to detect such attacks but lack the analysis for specific attacks. Machine learning (ML) algorithms are the ones used to detect such attacks and provide accurate results. This work is focused on detecting and mitigating black hole attack, gray hole attack, and flooding attack. Initially, performance of different ML algorithms in detecting these attacks is evaluated then later an algorithm for mitigating such attacks is proposed. WSN-DS dataset is obtained by inducing a malicious node acting as gray hole and black hole attacks. This dataset is given as the input of classification models with nine different ratios for training and testing (90:10 to 10:90). Performance measures such as accuracy and execution time for various classifiers were analyzed for each attack. From the results, it is observed that adaboost classifiers have the highest average accuracy of 97.97 and 94.48% in black hole attack and flooding, whereas random forest achieved 98.26% for gray hole attack. Similarly, in execution time, random forest took the least time of 0.081 µs. The analysis of ML algorithms is carried out using Jupyter notebook, simulation and mitigation of attacks is carried out in network simulator 2 (NS2).KeywordsWireless sensor networksDenial of servicesBlack hole attackGray hole attackFloodingMachine learningNS2