Development of Hard Drive Failure Prediction Model for Cloud Platform Using Intelligent Techniques

Abstract

Disk failures in cloud platforms remain a critical reliability concern because they can cause severe data loss, service downtime, and financial losses. This study aims to develop an Adaptive Neuro-Fuzzy Inference System (ANFIS)-based hard drive failure prediction model, investigate the impact of selected Self-Monitoring, Analysis, and Reporting Technology (SMART) attributes on predictive performance, and evaluate ANFIS against existing prediction techniques. A quantitative predictive modeling approach was employed using Backblaze SMART telemetry data, with Recursive Feature Elimination (RFE) applied for feature selection. Eight critical SMART attributes were selected, including reallocated sector count (SMART 5), seek-error rate (SMART 7), and temperature (SMART 231). The proposed ANFIS model achieved 89.4% accuracy, 91.2% precision, 87.8% recall, and an area under the curve (AUC) of 0.934. Comparative results show that ANFIS outperformed Random Forest, Gradient Boosting, Neural Networks, and Support Vector Machines (SVMs) in predictive performance. The study concludes that integrating ANFIS with RFE provides an effective and interpretable approach for hard drive failure prediction in cloud computing environments. These findings contribute to intelligent predictive maintenance research by demonstrating the value of neuro-fuzzy modeling for improving disk failure detection, supporting proactive maintenance, reducing downtime, and enhancing operational reliability in large-scale cloud platforms.