POSTER: Feature Selection to Optimize DoS Detection in Wireless Sensor Networks
Mousa Al-Akhras, Abdulaziz I. Al-Issa, Mohammed S. Alsahli, Mohammed Alawairdhi · 2020
Denial of Service is the most common attack in Wireless Sensor Networks (WSNs). Decision Trees (DT) and Artificial Neural Network (ANN) are used to detect attackers' signature. The most relevant features and the most dangerous attacks are selected from the WSN dataset. Feature selection reduces the time needed to learn the attackers' signature and improves the speed of detection as it relies on a reduced set of features. Results show that both DT and ANN achieved the best results when only the relevant features are used and the most dangerous attacks are considered. DT achieved 99.83% classification accuracy, 0.998 True Positive (TP) rate and 0.014 False Positive (FP) rate. The best DT has 463 nodes and 232 leaves. The best ANN model achieves 99.76% accuracy, 0.998 TP, 0.004 FP, and it has 28 neurons which is considerably smaller than the best DT.