Effective Feature Extraction Framework to Improve Network Intrusion Detection System

A network intrusion detection system (NIDS) is a crucial component of a robust cybersecurity strategy. Its primary purpose is to continuously monitor network traffic and detect suspicious or malicious activity that could be indicative of a cyberattack or unauthorized network access. The effectiveness of NIDS depends heavily on the techniques we use to boost the classification accuracy of intrusion and minimize the computational difficulty while performing training and testing. The high volume of network traffic combined with its large number of features will increase the classification time. With the recent emergence of deep learning techniques, scientists have shown interest in learning dataset features, followed by the classification of intrusions. This study offers a novel method for extracting high-dimensional features from input data by employing a stacked sparse autoencoder. Simple machine learning models are then built using the remaining low-dimensionality features. Simulations were conducted, and the efficacy of binary and multiclass classifications was verified. The proposed method exceeds most of the other existing approaches in terms of performance.