TY - JOUR
AU - Li, X.
AB - This paper presents an optimized machine learning-based approach for hard disk failure prediction, specifically tailored for shipboard systems. In shipboard systems, unexpected hard disk failures can disrupt operations and lead to critical data loss; therefore, reliable failure prediction is essential. To address the impact of dataset imbalance on predictive performance, we employed the Synthetic Minority Over-sampling Technique (SMOTE) alongside a fault-backtracking method to transform the imbalanced dataset into a balanced one, thereby enhancing the algorithm's predictive accuracy. Four machine learning algorithms were utilized in this study: K-Nearest Neighbors (KNN), Logistic Regression (LR), Extreme Gradient Boosting (XGBoost), and Random Forest (RF). Experimental results demonstrated that the RF algorithm achieved an accuracy of 97.1%, outperforming the other algorithms across all evaluation metrics. The contribution of this work lies in providing a feasible solution for hard disk failure prediction in shipboard systems through effective data processing strategies and algorithm selection, ultimately enhancing shipboard safety and reliability.
TI - Employing machine learning paradigms for optimized hard disk failure prediction in shipboard systems
JF - Proceedings of SPIE
DO - 10.1117/12.3055308
DA - 2025-02-25
UR - https://www.deepdyve.com/lp/spie/employing-machine-learning-paradigms-for-optimized-hard-disk-failure-dnMOv7XM1h
SP - 1354205
EP - 1354205-6
VL - 13542
IS - 
DP - DeepDyve
ER -