A novel density estimation based intrusion detection technique with Pearson's divergence for Wireless Sensor Networks.

We present a novel technique to detect an intrusive attack that occurs in the network due to the presence of a compromised node. These intrusive attacks last for a long time in the network due to the existence of compromised nodes this also affects the sensor reading. As the time span of the attack in longer in the network, it affects the system and can cause a system failure. Hence, we propose a technique that uses the combination of multi-varying kernel density estimation with distributed computing. This combination analyzes the individual probability of the existence of data and calculates the global value of the Probability Density Function (PDFs). Pearson's divergence (PE) is applied for efficient in-network detection and estimation of intrusion at low False Positive Rate (FPRs). The approximation of PE divergence is carried out using different techniques of distributed computing. The value of PDFs is calculated for a successive period of time in order to provide efficient performance. We also propose an entropy-based method that uses a centralized computing approach. Results obtained using PE divergence and entropy-based method are compared in order to judge the robustness. Finally, the proposed algorithms are evaluated using real-world based datasets, and the results are compared using Accuracy and FPRs.