Hard Disk Drive Failure Prediction with Explainable Time-series Characteristics Tree and Large Language Model Hyperparameter Optimizer

Predictive maintenance of hard disk drives is crucial for protecting data assets across industries. Efficient maintenance planning requires real-time, accurate, and explainable failure prediction. Aiming at overcoming the trade-off between model accuracy and explainability, this work introduces a novel optimization approach for a tree-based explainable model. Our approach leverages the sophisticated reasoning capabilities of Large Language Models (LLMs) to identify the optimal hyperparameter configurations. The Chain-of-Thoughts (CoT) prompting strategy enables LLMs to generate and refine hyperparameter configurations iteratively, specifically targeting signal length and lead time for Self-Monitoring, Analysis, and Reporting Technology (SMART) time series segmentation and crucial tree model parameters like maximum depth and minimum sample splits. Experimental results demonstrate that the tree classifier using the LLM-based optimization method achieved an average performance that was 6.14% higher than that obtained through random search across 12 hyperparameter tuning experiments, showcasing the efficacy of LLMs in industrial optimization tasks.