Bayesian Optimization of XGBoost Model for Machine Failure Prediction and Model Interpretability Analysis

In the field of industrial manufacturing, advance fault prediction of machines and equipment can greatly reduce costs, improve efficiency and production quality, this study constructs an XGBoost model to achieve the prediction of machine faults. In view of the high sensitivity of machine learning algorithms to hyperparameters, Bayesian optimization theory is used to automatically optimize the parameters of the XGBoost model to achieve efficient prediction. Meanwhile, three machine learning algorithms, namely Logistic Regression, Random Forest, and Support Vector Machine, are utilized to construct the fault prediction model respectively. The experimental results show that when using the same training and test data, the BO-XGBoost model correctly predicts the number of faulty machines as the first of the four models, and the accuracy of predicting the fault class reaches 0.93, which shows that it has high accuracy. In addition, this study utilizes the SHAP model for interpretability analysis of the BO-XGBoost model, revealing the significant positive effects of the features of organic concentration (VOC), ultrasonic sensor data (USS), and air quality (AQ) on the predicted values of the model, with the importance of the VOC feature being the most prominent in particular. The method proposed in this study has significant practical application value and provides strong technical support for industrial safety production.