EXPLAINABLE DEEP LEARNING-BASED INTRUSION DETECTION SYSTEM FOR WIRELESS SENSOR NETWORKS WITH REAL-TIME EDGE DEPLOYMENT

Abstract: Wireless Sensor Networks are being used more and more in areas like healthcare and smart cities. This makes them a big target for cyberattacks. There are some problems with the ways we currently detect these attacks. Firstly the systems have a time detecting the attacks that do not happen very often. Secondly it is hard to understand how these systems make their decisions. Lastly it is difficult to use these systems in time. This paper talks about a system for detecting cyberattacks in Wireless Sensor Networks that use Wi-Fi. The system uses a kind of computer program called a Convolutional Neural Network. This program was trained using a lot of data from a dataset called AWID-CLS-R. The people who made this system did three things to make it better. They made sure the system had an amount of examples of each type of traffic. They also used a tool called SHAP GradientExplainer to understand which features of the traffic are most important for detecting each type of attack. Lastly, they made a dashboard that can be used in time to detect attacks.The system was. It worked very well. It was able to identify attacks 99.3% of the time. The system is also very small so it can be used on devices that do not have a lot of power like the Raspberry Pi 4.The people who made this system found out that some features of the traffic are more important than others for detecting types of attacks. For example the feature called wlan.fc.order is very important for detecting Flooding and Impersonation attacks. The feature called wlan.seq is very important for detecting Injection attacks. Overall this system is an improvement over the systems that are currently being used. It can detect attacks well it can explain how it makes its decisions and it can be used in real time. Wireless Sensor Networks are used in areas, including healthcare, smart cities and industrial monitoring and this system can help keep them safe, from cyberattacks including Flooding, Impersonation and Injection.

Index Terms - Intrusion Detection System, Wireless Sensor Network, Convolutional Neural Network, Deep Learning, SHAP Explainability, Class Imbalance, Balanced Under sampling, AWID-CLS-R, IEEE 802.11, Wi-Fi Security, Real- Time Monitoring, Feature Selection, Network Traffic Classification, Edge Deployment.