A HYBRID DEEP LEARNING AND GENETIC ALGORITHM MODEL FOR EXPLAINABLE NETWORK TRAFFIC CLASSIFICATION

Abstract

This thesis presents an explainable deep learning model for network traffic classification using a genetic algorithm. A ResNet-based classifier is combined with a GA-driven dominant feature selection method to enhance interpretability and optimize accuracy. Experiments on real-world encrypted traffic datasets achieved 97.24% accuracy while identifying critical statistical features. The model effectively balances accuracy, simplicity, and transparency, contributing to the advancement of explainable artificial intelligence in network security analysis.