ERF-XGB: An Edge-IoT-Based Explainable Model for Predictive Maintenance

As the number of Internet of Things edge devices in smart factories increasing, it is crucial to predict the lifetime of the equipment to keep the production running normally. Although predictive maintenance based on machine learning achieve a better performance, they still face the challenge of black box and time-efficient. This paper propose a stacking-based model called ERF-XGB for predictive maintenance in the edge computing environment. The model first performs initial prediction by integrating the Random Forest model and Extreme Gradient Boosting model, followed by further processing of the initial prediction results using the linear regression model to obtain the final prediction results. In addition, the model incorporates the Shapley Additive exPlanations method, which can enhance the interpretability of the model when performing predictive maintenance. An experimental evaluation of the Predictive and Health Management dataset shows that the ERF-XGB model has an RMSE of 18.271 and an MAE of 13.454, which are the two metrics that perform the best when compared to other comparison models, suggesting that the model has a better predictive performance. Meanwhile, the Shapley Additive exPlanations method visualizes the impact of each edge device on normal operation in the production process, facilitating precise equipment management and maintenance.