Optimizing Activation Functions for Intrusion Detection Systems Using Evolutionary Algorithms

This paper explores the importance of activation functions in enhancing the performance of intrusion detection systems (IDSs) using deep learning. Traditional activation functions such as ReLU, Sigmoid, and Tanh face issues like vanishing gradients and information loss. To address this, the authors propose a Genetic Algorithm (GA)-based framework to evolve new activation functions tailored for IDS. Tested on two datasets NSL-KDD and CICIDS 2017, these optimized functions improved detection accuracy by up to 3% and reduced false positives, demonstrating the effectiveness of evolutionary optimization to strengthen IDS in cybersecurity.