A genetic algorithm based auto-encoder based approach for intrusion detection system

The purpose of intrusion detection systems is to improve software and system security. The challenge of attack detection has already been addressed through supervised and unsupervised machine learning approaches. Although the previous methods lead to models that are very accurate on the observed samples, the new methods provide robust models on the unobserved samples. The accuracy of the new methods is equal to the accuracy of the observed samples. In this paper, we apply a deep neural network-based auto-encoder to the GA-AE-IDS (Automatic Encryption Detection System-Genetic Algorithm) intrusion detection system. Basically, the GA-AE-IDS is a light intrusion detection system that can be used online. In this paper, weighted datasets are processed in GA. First, auto-encoder neural networks are initialized; then GA-based weight optimization is performed on the system. The experimental results reflect that the proposed method can detect most attacks with good accuracy. It has also accelerated the training and testing process, and in dense samples has improved by about 4.9% compared to the base paper. Also, the AUC criterion in dense and sparse samples obtained from the proposed method has been improved by 5 and 8%, respectively.