Cloud computing infrastructures rely extensively on RAID-based storage systems to ensure data reliability and high availability. However, conventional RAID failure detection mechanisms are largely reactive, resulting in unexpected downtime and increased operational costs. This paper proposes an AI-driven predictive disk failure management framework for cloud-based RAID storage systems. The proposed approach analyzes SMART disk attributes and storage performance metrics to predict failures before they occur. A Random Forest classifier is trained using a SMART-based dataset augmented with simulated RAID failure scenarios and is evaluated against threshold-based monitoring and Support Vector Machine (SVM) baselines. Experimental results on 12,000 disk health records demonstrate that the proposed model achieves 94.2% accuracy, 93.5% precision, and 92.8% recall, significantly outperforming conventional approaches. The framework supports scalable cloud deployment and proactive alerting, thereby improving storage reliability and reducing downtime.