Intrusion detection system using machine learning and deep learning algorithms for NACOTS system

Abstract

The Internet continues to grow and spread across the globe, resulting in an enormous amount of data. Consequently, the number of cyberattacks grows daily. Intrusion Detection Systems (IDS) are indispensable security tools for detecting cyberattacks and malicious network traffic. Machine Learning and deep learning techniques are proposed to classify and analyse the network traffic content to identify abnormal activities and impending cyberattacks. Current intrusion detection techniques still face many obstacles, including a low detection rate, a high prediction latency, a high false alarm rate, and using an outdated dataset. This research aims to develop a fast and accurate algorithm for detecting cyberattacks. Dimensionality reduction is vital in the development of an intrusion detection system to reduce the complexity of the dataset and the inference time for faster detection of cyberattacks. The proposed feature selection is based on maximizing relevance and minimizing redundancy using Correlation Feature Selection (CFS), Mutual Information (MI), and Recursive Feature Elimination (RFE). The proposed feature selection algorithm has reduced the dataset's features from 78 to 25 features with remarkable results: an accuracy of 99.718% and 99.915% and FPR of 0.0929% and 0.0281% using CNN and XGBoost, respectively, and an inference time of 50.79μs and 4.78μs, respectively. Experiments were conducted using the CICIDS2017 dataset to evaluate the efficacy and efficiency of our developed IDS and the need to secure E-healthcare applications. Due to its low prediction latency, high accuracy, and high detection rate, the proposed IDS could be used as a server-side security layer for the server of NACOTS (Nanosystem for COVID-19 DNA/Antibodies On The Spot Test) application.