Enhanced Cyber Incident Detection and Prediction Using Machine Learning and SMOTE-Based Class Balancing
Abstract: Cybersecurity incidents continue to escalate in complexity and volume, necessitating advanced detection systems. This study investigates the application of machine learning models-specifically Decision Trees, Logistic Regression, Random Forest, Gradient Boosting, XGBoost, and LightGBM-to detect cyber threats. A dataset of 9.5 million entries with 45 features was used. Initial evaluation revealed XGBoost achieved the highest macro F1-score (0.91). To address class imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) was applied, improving XGBoost?s performance to 0.90 accuracy, 0.92 precision, recall, and F1-score. The findings support the integration of ensemble learning and data balancing for robust and scalable threat prediction. This research underscores the effectiveness of ensemble learning techniques, particularly XGBoost, in predicting and mitigating cyber threats, offering a scalable and efficient approach to cybersecurity incident detection.
Keywords: Cybersecurity, Machine Learning, XGBoost, SMOTE, Threat Detection
How to Cite?: Dr. S. Gnanamurthy, M. Arun Kumar Naik, "Enhanced Cyber Incident Detection and Prediction Using Machine Learning and SMOTE-Based Class Balancing", Volume 14 Issue 10, October 2025, International Journal of Science and Research (IJSR), Pages: 1284-1291, https://www.ijsr.net/getabstract.php?paperid=SR251024103823, DOI: https://dx.doi.org/10.21275/SR251024103823