Evaluation of Machine Learning Models Developed for the Detection of Adversarial Attacks on Wireless Network

Adversarial attack is a tactics employed by cybercriminal to deceive modern security systems and penetrate networks for cyber-attack. While several studies focused on learning threat patterns in datasets, there has been gap in their inability to detect unknown threat features. To solve this problem, this work focused on normal packet features as the vector of interest and then classified anything not normal packet as threat. The design of the study involved the collection of data from the networks of Nigerian Immigration Service (NIS), Enugu and Awka passport offices, considering metrics such as latency, throughput, packet loss, MTTR, error rate and MTTD. To improve the network security, several machine learning algorithms such as Artificial Neural Network (ANN), Decision Tree (DT), Random Forest (RF) and ensemble model were trained. Their results considering accuracy reported 93% for ANN, 81.87% for DT, 89.75% for RF and 97.39% for ensemble model. Precision of the models reported 90% for ANN, 80.78% for DT, and 98.46% for ensemble model. Recall recorded 90% for ANN, 83% for DT, 88.54% for RF and ensemble model recording 96.91%. F1-score reporting 91% for ANN, 84.16% for DT, 86.06% for RF and ensemble model recording 94.37% for f1-score.