Intelligent Artificial Intelligence-Powered Cyber Threat Detection System for Cloud Security Enhancement

Abstract

Network intrusion detection is a key part of keeping computer networks and the Internet of Things safe from cyber dangers. When it comes to decision-making, standard machine learning methods can only improve detection accuracy (ACC). This work introduces an explainable intrusion detection method that utilizes the CICIDS2017 dataset and depends on a Random Forest classifier (RF), LightGBM (Light Gradient Boosting Machine), and Voting Ensemble (VE). The data processing, feature selection, data balancing with SMOTE, and ensemble learning techniques of the proposed method are the foundation of intrusion detection ACC. At an ACC of 98.22%, the Voting Ensemble model outperforms the other models mentioned in the discussion (Multiple Baseline models), including the Autoencoder (AE), Decision Tree (DT), Deep Neural Network (DNN), and Logistic Regression (LR) classifiers. To further enhance model transparency and determine the most important network traffic features for attack prediction, SHAP-based explainable AI analysis is used. The proposed framework is able to identify influential traffic aspects that lead to attack prediction, subsequently leading to an accurate, scalable and trustworthy solution for current intrusion detection applications, thereby increasing the interpretability of cybersecurity.