From LSTM to Chronos: Evaluating Deep Learning Models for Predictive Maintenance

This study evaluates three approaches for predicting Remaining Useful Life (RUL) using the C-MAPSS FD001 dataset: Long Short-Term Memory (LSTM), XGBoost, and Chronos, a transformer-based time series model inspired by large language models. The LSTM model is enhanced with statistical feature selection to improve robustness and interpretability. Experimental results show that the optimized LSTM model achieves the best performance, with a Root Mean Squared Error (RMSE) of 14.32, outperforming both XGBoost and Chronos. These findings highlight the effectiveness of well-tuned recurrent models and suggest that larger architectures do not necessarily guarantee better predictive accuracy.