A Multi-Objective Approach to Improve Hyperparameters of CNN for Network Intrusion Detection Problem

This paper proposes and develops a method for improving hyperparameters in convolutional neural networks for network intrusion detection problems, based on swarm optimization algorithms and a multi-objective approach. Based on Pareto principles and machine learning model assessment measurements, the global optimal objective function aims at a balance between objectives: Accuracy, Precision, and Recall. The value of the objective functions is assessed based on the measurements achieved after each model training process with a specific set of hyperparameters (structures). The global objective function is used as a fitness function in the PSO algorithm. Use PSO to choose the best set of hyperparameters. The method proposed in the paper was tested on 2 datasets, in which the value of the global objective function of the two datasets reached 89.635% and 78.85%, and the improvement level of the global objective function corresponding to the two datasets reached 0.006% and 0.087%.