Linear Regression and Random Forest: Comparison on Industrial Data

Abstract

Industry 4.0 promotes the wide use of sensing devices in smart factories, generating massive operational data that requires reliable prediction tools for energy management and equipment maintenance. In particular, linear regression and random forest are two mainstream prediction algorithms frequently adopted for industrial data analysis. Existing research generally discusses these two models separately, while few comparative studies use unified industrial datasets to test their practical performance. This paper adopts open factory energy consumption records to carry out controlled comparative tests, and evaluates model performance through four universal statistical evaluation metrics. Analytical results indicate that linear regression provides fully interpretable prediction results yet only adapts to simple linear datasets. Random forest delivers stronger fitting capacity for non-linear and collinear industrial data, but its internal computing logic cannot be clearly explained. This study further summarizes applicable industrial scenarios for each algorithm and proposes a two-stage hybrid modeling scheme. This hybrid scheme not only balances interpretability and predictive accuracy, but also offers a practical reference for selecting or combining algorithms in real-world industrial data analysis tasks.