AI-powered threat detection system

In an era of growing digital interconnectedness, the threat landscape for networked systems has expanded rapidly, making traditional security mechanisms increasingly ineffective against sophisticated cyber-attacks. Intrusion Detection Systems (IDS) are crucial in identifying and mitigating such threats, but conventional rule-based approaches often fail to detect novel or evolving attack patterns. This paper proposes an AI-powered IDS framework that utilizes Random Forest and Decision Tree machine learning models for high-accuracy threat detection in real time. The models are trained on benchmark datasets, namely NSL-KDD and CICDDOS2019, both widely used in cybersecurity research. Preprocessing techniques such as one-hot encoding and robust feature scaling were applied to optimize learning. The trained models are then integrated into a web application built with Flask, providing users with a seamless interface to upload network traffic logs in CSV format and instantly receive predictions. The system also incorporates rule-based logic to categorize detected attacks into DoS, Probe, R2L, and U2R, enhancing interpretability. Evaluation results demonstrate that the Random Forest model achieved a classification accuracy of 99.36% and an F1-score of 0.9986, outperforming the Decision Tree model across all metrics. The application supports real-time traffic classification, returning predictions within seconds and displaying confusion matrices, precision, recall, and attack distributions through a clean, responsive UI. This research bridges the gap between theoretical machine learning models and their real-world application in cybersecurity, offering a scalable, accurate, and user-friendly solution for automated threat detection in both academic and professional environments.