Predicting Dew Point Temperatures: A Machine Learning Approach with SHAP Explanations

Abstract

In industrial production, controlling the dew point temperature is crucial for maintaining operational efficiency and product quality. This research aims to apply machine learning models to predict dew point temperatures and enhance model interpretability using SHapley Additive exPlanations (SHAP) to explain feature contributions. The data was collected from an Industrial Internet of Things (IIoT) system, and models evaluated include Linear Regression, Ridge Regression, Lasso Regression, Decision Trees, Random Forest, Gradient Boosting, XGBoost, Support Vector Regression (SVR), and k-Nearest Neighbors (KNN). The results indicated that the Random Forest model performed best, with the highest R² (0.94) and the lowest RMSE (0.82). Other well-performing models include Gradient Boosting, with an R² of 0.93 and an RMSE of 0.86, and XGBoost, with an R² of 0.93 and an RMSE of 0.87. For model interpretability, the real-time power consumption of the system ("ACTIVE POWER"), supply air temperature ("TEMP IN"), and supply air humidity ("HUM IN") were identified as important factors influencing the predictions. The SHAP analysis provided local and global insights into feature importance, enabling more informed decision-making in dew point control. These findings demonstrate the potential of integrating machine learning and explainable AI in industrial applications to advance operational strategies and safety measures.