An Ensemble Framework for Predictive Maintenance of WaterPumps

This research work introduces an ensemble framework for predictive maintenance of water pumps and empirically shows its efficacy. This study has examined the effects of feature extraction methods and class balancing procedures on the effectiveness of ensemble model. The results show that the oversampling improves accuracy with a considerable margin and achieves a remarkable 99.6% accuracy and MAE of 0.003. Additionally, the inclusion of all characteristics outperforms feature extraction with Principal Component Analysis (PCA), highlighting its contribution in improving the prediction accuracy. By achieving minimum MAE of 0.003 and accuracy of 99.64%, comparative analyses against base models and other ensemble techniques further confirm that the proposed ensemble model is the most accurate and effective solution. These results demonstrate ensemble approaches' efficacy and effectiveness to advance the preventive maintenance system employed in water pump systems.