A Network Intrusion Detection Framework on Sparse Deep Denoising Autoencoder for Dimensionality Reduction

Abstract

"Network security" is currently among the crucial areas of computer science. Due to the proliferation of IoT tools and "peer-to-peer" systems, the necessity to mitigate safety concerns is prominent. Intrusion detection in network systems is employed to analyse network packets for malicious activity. Different kinds of attacks, such as "Denial of Service", "Probe", "Remote-to-Local", and "User-to-Root" are instances of unexpected behaviour. Once attacks are spotted, customized alerts can be sent to the personnel responsible. In numerous application fields of information security, intrusion detection is a necessity in the modern era. In this study, we examined the usefulness of several autoencoder types in identifying network intrusions. In this research, a system for network intrusion detection based on Sparse Deep Denoising Autoencoder for dimension reduction was established. If the autoencoder is trained on regular network information, reconstruction error (the dissimilarity between the original and recreated data) is used to identify anomalies. Intrusion Detection System Development utilized standard datasets such as KDDCup99, NSL-KDD, UNSW-NB15, and NMITIDS. The effectiveness of four distinct autoencoders was analysed to identify network attacks. We were able to achieve over 96% accuracy with only reconstruction error by using a sparse deep denoising autoencoder. The primary research aim was to enhance the performance of the network by achieving high intrusion detection precision.