A Machine Learning Framework for Cyber Risk Assessment in Cloud-Hosted Critical Data Infrastructure

Abstract

Cloud computing has rapidly expanded to offer scalability and flexibility to current organizations, but it has also posed a new era of challenging yet dynamic security threats. With cyber threats evolving to new levels of sophistication and magnitude, organizations are finding it difficult to manage cybersecurity risks. Cybersecurity risks to the cloud-based critical infrastructure are highly sophisticated and can exact profound risks on data integrity, financial stability and operational continuity. These dynamic threats tend to be hard to foresee and thus prioritized through the conventional means of risk assessments. The research work will describe a powerful machine learning system that performs well on cyber risk to include supervised classification methods, unsupervised clustering, and anomaly detection. The steps will include preprocessing a large general global cyber threat dataset, training models with Random Forest, XGBoost, and NGBoost, model performance estimation with accuracy, precision, recall, F1-score, ROC-AUC while conducting clustering and anomaly analysis to gain further insight. Findings indicate NGBoost is the most accurately predictive (98%) and has few misclassifications, clustering reveals interesting incident patterns, and anomaly detection pinpoints high-risk sectors. The framework demonstrates especially high accuracy, interpretability, and actionable insights compared to benchmark models (CNN and KNN), which makes it a scalable and reliable proactive tool of cyber risk management. The contribution of the work is combining sophisticated feature engineering, model tuning, and multi-level analysis effectively to increase the practical cybersecurity resilience.