AI-Driven Cyber Threat Intelligence for Predictive Attack Detection and Automated Security Response

Abstract

This study aimed to develop an AI-driven approach for predictive cyber attack detection using labeled network-traffic data and to evaluate machine-learning techniques for distinguishing benign from malicious activity. The CIC-IDS2017 dataset was used as the primary data source for the experimental analysis. Logistic Regression, Random Forest, XGBoost, and a Multilayer Perceptron were evaluated within a common machine-learning framework for binary classification of benign and malicious network activity. A feature-selection analysis was also conducted to examine whether a reduced set of informative network-flow variables could maintain effective detection performance. The comparative analysis identified XGBoost as the strongest-performing model among the evaluated approaches, demonstrating highly consistent classification performance and very limited misclassification between benign and malicious traffic. The feature-selection experiment showed that the reduced-feature configuration maintained strong attack-detection capability, with only a modest decline in performance compared with the complete feature representation. The findings indicate that ensemble-based learning can effectively identify discriminative patterns in structured network-flow data. The study provides empirical support for AI/ML-based predictive cyber attack detection and feature reduction in cybersecurity applications. Future research should validate the framework using independent datasets, diverse attack scenarios, realistic network environments, and an experimentally evaluated automated security response mechanism.