Intrusion detection system for wireless sensor network

Abstract

Wireless sensor networks (WSNs) have numerous application in almost every domain. WSNs are composed of cheap tiny devices that are deployed in open and unsafe environments making them exposed to all sorts of attacks. The security of such networks is of great importance. Hence, securely operating WSNs, any intrusions should be recognized before attackers can harm the network. Among the IDS solution, Machine learning-based IDSs have proved their e ciency. However, Machine learning-based IDS for WSN needs a set of features to characterize attacks while respecting WSN speci cities.In this work, we investigate how to sub-optimally characterize attacks in WSN. Based on 6,568 deployed simulations, we have got a large dataset of 94,426 records that captures di erent behaviors at the connection level. The targeted behaviors are Normal,Blackhole, Hello-Flood, and DoS. Based on classi cation performances, our results prove that at the connection level, attacks can be characterized by just four attributes. Using Random Forest classi er, we reached 91:8% of precision in the low-loaded scenario and 78:1% in the high-loaded scenarios.