Prediction of Hard Disk Remaining Useful Life Based on a Cross-Validation Accuracy Weighted Probability Ensemble Model

As one of the most widely used storage devices today, hard drives are efficient and convenient, but their failure can cause significant losses. Therefore, early warning of hard drive failures, facilitating timely backup and transfer of stored content, is crucial to minimizing these losses. In recent years, research on hard drive failure prediction has been emerging continuously. However, many models suffer from short prediction times spans and low accuracy, which can lead to insufficient time for data backup. To address this issue, this paper proposes a Cross- Validation Accuracy Weighted Probability Ensemble model (CAWPE) to extend the failure detection cycle. By in-corporating a softmax function in the model's output layer, the interpretability of the model's output is enhanced. Experimental results show that the improved ensemble learning model significantly outperforms baseline models, including BP neural networks, random forests, and LSTM, on datasets of various hard drive models.