TabNet-Based Multi-Class Cyber Security Threat Classification Using Attention-Based Deep Learning
Main Article Content
Abstract
The growing complexity and the number of cyberattacks have made it imperative to have smart intrusion detection system that can detect multiple network threats in real-time with accuracy. Traditional machine learning techniques are typically highly complex and intricate, and require significant feature engineering, and are usually unable to model the high-dimensional and complex network traffic characteristics. This research presents a deep learning-based framework for cyber security threat classification based on TabNet multi-class intrusion detection model. By incorporating a sequential attention mechanism to automatically select features that are most informative, the proposed model not only achieves state-of-the-art classification results but also introduces intrinsic model interpretability, thus addressing the issue of model explainability. Experimental evaluation was performed on the CIC-IDS2017 benchmark dataset that consists of more than 2.23 million preprocessed network flow instances of benign traffic and various categories of attacks. To ensure the robustness of model training and evaluation, the dataset was preprocessed by removing data duplication, encoding categorical labels, standardizing features, and stratifying the data. For the proposed TabNet model, the overall classification accuracy was 97.92%, which is very high, with weighted precision, recall, F1-score and weighted ROC-AUC of 97.69%, 97.92%, 97.64% and 99.56%, respectively, showing excellent discrimination capability for multiple cyber threat classes. In addition, feature importance analysis revealed the role of important network traffic characteristics in the classification process, making the models more transparent and suitable for explainable artificial intelligence (XAI) in the field of cybersecurity. The experimental results show that the proposed framework is an accurate, scalable, and interpretable approach to intelligent cyber threat classification and has good potential for application in enterprise intrusion detection systems, cloud computing and next generation cyber defense platforms.