Performance Analysis of Machine Learning Algorithms for Intrusion Detection in Wireless Sensor Networks

The widespread adoption of wireless sensor networks has accompanied technological advancements. The inherent sensitivity of wireless sensor networks necessitates robust protection against potential cyber-attacks. Numerous intrusion detection methods are under scrutiny to safeguard services and infrastructures of wireless sensor networks. This study uses machine learning techniques to detect Denial of Service (DoS) attacks in wireless sensor network data. The research investigates the impact of unbalanced distribution of traffic within wireless sensor networks on intrusion detection. Experimentation was conducted on the WSN-DS dataset, utilizing Logistic Regression, Decision Tree, Bernoulli Naive Bayes, K-Nearest Neighbor, AdaBoost, Gradient Boosting, and CatBoost algorithms for binary classification. Data balancing was achieved by applying SMOTE oversampling and RandomUnderSampler undersampling methods. The comparative assessment of model performances was conducted using metrics such as Accuracy, Precision, Recall, and F1-Score. Furthermore, the study examined the confusion matrix and ROC curve. Results indicated that data balancing methods decreased accuracy rates but effectively reduced misclassification of DoS attacks.