How Machine Learning Optimizes Data Center Disk Health and Power Efficiency

This study presents a machine learning framework that improves data center efficiency through selective disk scrubbing. By applying Mondrian conformal prediction to assign health scores and prioritize failing drives, the model reduces unnecessary maintenance operations—scrubbing only 22.7% of disks—saving significant energy and improving reliability. Tested on Baidu’s open-source dataset, this approach demonstrates how predictive modeling can optimize large-scale storage systems while minimizing operational costs.