Cost-Sensitive Bootstrapped Weighted Random Forest for DoS attack Detection in Wireless Sensor Networks

Security of Wireless Sensor Networks is vital as this class of networks is increasingly being used for mission-critical applications, surveillance, military and disaster management monitoring. Machine learning algorithms are used nowadays in Intrusion Detection Systems that form the first line of defense against security attacks. This work proposes a cost-sensitive machine learning-based classifier trained on the WSN-DS dataset [13] comprising flooding, TDMA/Scheduling, Black-hole, and Grey-hole attack samples. Our proposed algorithm handles the imbalanced nature of the dataset efficiently without relying on resampling techniques. Other techniques for handling imbalance can induce extra computational processing that is unsuitable for battery-powered and resource-constrained sensor networks. Given this, we propose Cost-Sensitive Bootstrapped Weighted Random Forest (CSBW-Random Forest), which demonstrated superior performance over existing works. Our method gives the accuracy, precision, recall, and F1-score of 0.997, and per-class performance scores are also in the range of 0.95 to 0.99, which is significantly better than existing literature. The analysis also indicates that the proposed work is giving a better true positive rate (0.979), false-positive rate (0.003), false-negative rate (0.020) than related works. Experimentation with proportionately increasing data samples also validates the higher performance of our model.