Genetic Algorithm–Based Hyperparameter Optimization for 1D-CNN NIDS

This paper investigates Genetic Algorithm (GA)–driven hyperparameter optimization for one-dimensional Convolutional Neural Networks (1D-CNNs) in network intrusion detection systems (NIDS). Building on prior evidence that evolutionary search markedly improves deep models for NIDS, we optimize nine key hyperparameters (filters, kernel size, pooling size, dense depth/width, dropout, learning rate, batch size, and epochs) and train/evaluate on UNSW-NB15, CIC-IDS2017, and NSL-KDD using standard metrics: accuracy, loss, precision, recall, and F1-score. Our GA framework encodes hyperparameters as chromosomes and evolves candidates via selection, crossover, and mutation to maximize validation performance. Experiments show that GA-tuned 1D-CNNs consistently outperform non-optimized baselines across datasets; on UNSW-NB15, for example, GA attains ~99.31% accuracy, aligning with the best reported performance for GA-optimized 1D-CNNs. Results highlight that GA delivers robust gains with practical compute budgets while preserving strong precision-recall trade-offs, and that effectiveness can vary by dataset characteristics. Overall, GA-based hyperparameter optimization offers a simple, reproducible path to higher accuracy and reliability in deep learning–based NIDS, advancing the development of adaptable intrusion-detection solutions for evolving cyber threats.