A steel enterprise is currently trying to avoid the presence of surplus materials since they can greatly increase its operational cost. The complicated production process of steel products makes it difficult to find the causes of surplus materials. In this work, we propose a surplus material prediction problem and solve it based on statistical analysis and machine learning methods. In the concerned problem, we predict whether there are surplus materials under a given group of production parameters. The dataset used in this work is from a real-world three-month steel coil production process. First, data cleaning is conducted to standardize the industrial dataset. Then, the production parameters highly correlated with surplus material prediction results are selected by a series of feature selection methods. Finally, two prediction models based on extreme gradient boosting and logistic regression are presented according to the selected features. The experimental results reveal that the proposed prediction models have similar effectiveness. A visible regression function makes the logistic regression method more suitable for practical application.