Explainable AI for Intrusion Detection Systems

Abstract

The recent advancements in networks facilitates faster communication to any part of the world. The widespread adoption of Internet of Things in daily life applications proposes networking of gadgets. With the applications of Network through interconnection being increased, the difficulty in maintaining a secure network state becomes a challenge. Intrusion Protection Systems and Intrusion Detection Systems are two widely used tools in network security maintenance. Anomaly based IDS designed with the help of AI, ML and DL algorithms is observed to be more efficient than conventional signature based systems in the literature. Even though the reported accuracy of IDS in all the literature so far is sufficiently high, false alarms raised by the system is a major issue. The lack of explainability in the designed classifier behaviour is an important reason which makes it inevitable to avoid raising false alarms. This paper proposes an Interpretable A-IDS using XAI techniques. LIME and SHAP explanations are easily Interpretable, reducing the chances of raising false alarms.