Intelligent Intrusion Detection Framework Using RuleGuided and Deep Learning Techniques

Abstract

This work introduces an intelligent intrusion detection framework that combines ruleguided preprocessing and advanced machine learning techniques to accurately identify and classify network attacks using the NSL-KDD dataset. The system provides an interactive interface covering all essential steps, including dataset upload, preprocessing, model training, and performance evaluation. During preprocessing, categorical attack labels are transformed into numerical identifiers to make the data compatible with machine learning algorithms, and the dataset is partitioned into training and testing sets for unbiased evaluation. Four algorithms—Support Vector Machine (SVM), Random Forest, Deep Neural Network (DNN), and Extreme Learning Machine (ELM)— are implemented and compared, with their predictive performance measured in terms of accuracy. Experimental analysis shows that SVM and Random Forest deliver moderate classification results, whereas the DNN achieves lower accuracy by effectively capturing complex, nonlinear patterns in network traffic. The ELM demonstrates the advantage of rapid training, offering a fast yet reasonably accurate alternative. Visualization tools within the interface provide clear comparative insights into model performance, highlighting the ELM as the most effective method. Overall, the framework delivers a structured and efficient solution for intrusion detection research, demonstrating the benefits of deep learning approaches and laying the groundwork for scalable, real-time cybersecurity monitoring systems.