Abstract

Cloud computing has transformed the delivery of computing resources by providing scalable, on‑demand access to processing, storage, and software services. Despite the benefits in cost and flexibility, cloud adoption poses significant security challenges due to the dynamic, distributed, and multi‑tenant nature of cloud environments. Traditional security risk assessment approaches struggle to keep pace with evolving threats such as intrusion attempts, data breaches, misconfigurations, and advanced persistent threats. Machine learning (ML) techniques have emerged as powerful tools for proactive security management, enabling automated risk assessment, anomaly detection, and threat prediction by learning patterns from historical data. By integrating supervised, unsupervised, and deep learning models, cloud security systems can classify activities as normal or malicious, identify vulnerable configurations, and forecast potential security incidents. This paper explores how machine learning techniques enhance cloud security risk assessment and threat prediction, synthesizing research from foundational work through 2021. It examines data collection strategies, feature engineering, model training, evaluation metrics, and deployment challenges. Findings highlight that ML‑based security systems improve detection accuracy, reduce response time, and support adaptive threat mitigation, yet they also face challenges such as data imbalance, feature drift, interpretability, and computational overhead. Recommendations and future research directions include hybrid modeling, adversarial robustness, and explainability for ML‑based cloud security.