Research on Steel Product Quality Prediction Based on Random Forest and Gradient Boosting Machine

This paper investigates the prediction of steel product quality using Random Forest and Gradient Boosting Machine methods. To enhance data quality and the predictive accuracy of models, this study analyzes the relationships between variables through Kendall correlation coefficients, providing an important foundation for further analysis. In terms of model construction, the paper first uses the Random Forest model for feature importance analysis, finding that carbon content, soaking furnace temperature, and slow cooling furnace temperature significantly affect the hardness parameter. Subsequently, feature engineering methods are employed to calculate various statistical properties of the data, constructing a new feature matrix to improve the predictive performance of the model. Ultimately, the paper constructs multiple predictive models, including multiple linear regression, ridge regression, gradient boosting regression, and BP Neural Network, and selects the optimal model for efficient and accurate prediction of steel product quality through systematic evaluation of these models' performance. The model evaluation results show that the Gradient Boosting Regression model performs the best in all assessment metrics, demonstrating high predictive accuracy and explanatory power. This study provides data-driven model support for online inspection of steel product quality.