Hard Disk Failure Prediction Based on Blending Ensemble Learning

As the most widely used storage device today, hard disks are efficient and convenient, but damage incurred in event of a failure can be very significant. Therefore, early warnings before disk failure, allowing stored content to backed up transferred advance, reduce many losses. In recent years, an endless stream research on prediction has emerged. The detection accuracy various methods, from basic machine learning models, such as decision trees random forests, deep BP neural networks recurrent networks, also been improving. this paper, based idea blending ensemble learning, novel method combining algorithms is proposed publicly available BackBlaze datasets. experiment conducted only with S.M.A.R.T., that is, learned characteristics collected by self-monitoring analysis reporting technology, which internally counted during operation disk. experimental results show model able outperform other independent models terms evaluation criterion Matthews correlation coefficient. Additionally, through multiple types disks, high performance found, solves problem low robustness generalization traditional methods proves effectiveness universality method.