Performance Evaluation of a Genetic Neuro-Fuzzy Intrusion Detection System across Multiple Datasets

Abstract

The paper introduces an IDS that combines a genetic-algorithm feature selector with an Adaptive Neuro-Fuzzy Inference System classifier. A genetic algorithm, one of the most prominent heuristic optimization methods, is utilized to select a set of optimal features to serve as inputs to the IDS. The performance of this hybrid approach is rigorously compared with the widely adopted open-source Snort system using several standard benchmark datasets, including KDDCup99, NSL-KDD, UNSW-NB15, Bot-IoT, and CSE-CIC-IDS2018. The primary objective is to create a system capable of learning and detecting previously unknown attacks by harnessing the strengths of neural networks and fuzzy logic, thereby minimizing erroneous classifications—whether considering benign data as malicious or vice versa. The model is trained and tested on five public datasets and benchmarked against Snort. Across all datasets the GA-ANFIS variant attains higher accuracy (≈99%) and markedly lower false-positive rates (<0.3%) than Snort, implying better adaptability to diverse attack patterns. The empirical results demonstrate that the proposed system exhibits substantial potential in enhancing detection accuracy and adaptability to emerging threats.