Abstract
As network security continues to emerge as a fundamental concern, significant advancements in design have transpired over recent years. Amid various approaches, Intrusion Detection Systems (IDS) have garnered substantial attention. It appears that the complex nature and uncertainty inherent in security breaches render fuzzy techniques well-suited for such systems. Consequently, this research endeavours to harness the potential of the Adaptive Neuro-Fuzzy Inference System (ANFIS) as a classifier for distinguishing network instances into malicious categories (Probe, DoS, U2R, R2L) and normal behaviour. This classification is executed on the KDD99 database and compared against other machine learning models like Decision Tree and Multilayer Perceptron. Gauss-ian, Triangular, bell-shaped, sigmoidal, membership functions are investigated and gaussian is found to be the best for this problem. Moreover, the empirical findings emphasize ANFIS's superior performance, capitalizing on the strengths of both Artificial Neural Networks (ANN) and fuzzy reasoning systems, encompassing the ability to comprehend nonlinear interaction patterns, adaptability, and swift learning capabilities.