Design and Implementation of Network Intrusion Detection System based on Machine Learning
Rongguo Fu
Abstract
Traditional rule or signature-based Intrusion Detection Systems (IDS) often struggle to effectively identify and defend against new, complex, and evolving cyber attacks. Therefore, this study explores machine learning (ML)-based network intrusion detection technologies to enhance the accuracy and flexibility of detection. Methodologically, this paper adopts Deep Neural Networks (DNN) as the core algorithm and combines it with the Random Forest algorithm from ensemble learning to construct a hybrid model that improves the performance and accuracy of network intrusion detection. The system design employs a layered architecture, encompassing five levels: data collection, data preprocessing, feature extraction, intrusion detection, and alerting and response, ensuring the system's scalability, maintainability, and efficiency. The research results indicate that the hybrid model of DNN and Random Forest achieved an accuracy rate of 98.5% on the test set, significantly higher than models using DNN or Random Forest alone. The system can complete real-time classification of network traffic within milliseconds, meeting the real-time requirements of IDS. Moreover, the model demonstrates strong robustness in the presence of noisy data and unknown attack types, maintaining a high detection rate while reducing the false positive rate. ML-based network IDS showcases its potential in addressing complex network threats, promising to drive technological innovation and development in the field of cybersecurity. This study successfully designs and implements an ML-based network IDS that employs an ensemble learning approach combining DNN and Random Forest, significantly improving the accuracy and robustness of network attack detection. The test set accuracy reached 98.5%, and the system meets real-time requirements, with an average processing time of 50 milliseconds.