Optimizing Fast Learning Network Using Fuzzy Adaptive Metaheuristic Algorithms for Robust Intrusion Detection

In today’s digital era, the prevalence of cyberattacks requires robust security measures to keep sensitive and confidential information and infrastructure safe and secure. The intrusion detection system (IDS) plays the most vital role in this aspect by serving as a first‐line defense against these malicious activities and potential breaches. This paper presents a hybrid approach to improve the security and safety of networks through an efficient IDS. The proposed approach for the new IDS optimizes the fast learning network (FLN) through a fuzzy adaptive metaheuristic algorithm. A Mamdani‐based fuzzy adaptive strategy (FAS) is incorporated into four metaheuristic algorithms: the equilibrium optimizer (EO), growth optimizer (GO), sine cosine algorithm (SCA), and arithmetic optimization algorithm (AOA). The FAS dynamically balances the exploration and exploitation operations of these algorithms to enhance their performance. FAS is a two‐input, one‐output fuzzy technique that selects the best search operator during the optimization process. The proposed hybrid models that combine FLN with the fuzzy‐based implementations of the four stated metaheuristic algorithms are trained and tested on benchmark intrusion datasets known as NSL‐KDD and CIC‐IDS2017. Based on the results of our experiments, the fuzzy‐based optimized FLN models outperform the FLN models optimized using the four fundamental metaheuristic algorithms in terms of testing accuracy and training time. Moreover, fuzzy adaptive AOA (FAAOA)–based FLN outperformed all competing models by generating the best overall accuracies as well as by taking less training time for different numbers of neurons in the hidden layer of FLN. Finally, all proposed fuzzy optimized FLN‐based IDS models exhibited resilience against the fast gradient sign method (FGSM) and projected gradient descent (PGD) adversarial attacks.