AN ENHANCED INTRUSION DETECTION FRAMEWORK USING SMOTETOMEK AND RANDOM FOREST FOR WIRELESS SENSOR NETWORKS

Abstract

The widespread use of Wireless Sensor Networks has become an unavoidable necessity in many critical situations. Simultaneously, making them prime targets for network attacks such as Grayhole, Blackhole, TDMA, and Flooding attacks. Traditional Intrusion Detection Systems often struggle with different problems, such as class imbalance, that lead to poor attack detection accuracy, especially for minority class attacks. To address this issue, this study proposes an advanced network threat detection system framework integrating SMOTETomek with Random Forest for efficient performance in the analysis and classification of network threats. The study uses the WSN-DS dataset for training and evaluation of the methods. The SMOTETomek technique is employed to mitigate class imbalance by oversampling minority attack instances and removing noisy overlapping samples, ensuring a more balanced and representative dataset. The Random Forest technique, which is more familiar for its ensemble learning capabilities, is used for classification. The experimental results demonstrate a high performance, where the overall accuracy is 99.38%, with per-class F1-scores exceeding 97.62%, outperforming many existing IDS techniques. Therefore, the proposed framework can be a more advanced and effective solution in enhancing cybersecurity in WSNs.