Intrusion Detection System Using Hybrid Model of Denoising Autoencoder and Ladder Variational Autoencoder

In the realm of cyber-security, the identification and mitigation of network intrusions remain critical challenges. The deep learning based approach for intrusion detection was introduced on reconstructing network traffic dataset using 3 autoencoder architectures and classification using Convolutional Neural Networks (CNN) and Gated Recurrent Units (GRU). The proposed technique is assessed using 10% of the CIC-IDS-2017 dataset. Three autoencoder architectures (GDAE, LVAE, and Gaussian Noise-injected LVAE) are used to remove noise and extract relevant features. This technique successfully identifies network intrusions, with an overall performance of 99% across all assessment measures.