Abstract
This research presents an enhanced predictive model for green sand casting, designed to tackle the nonlinear complexities arising from interdependent process parameters. Casting defects substantially affect product quality and rejection rates, making accurate prediction vital. To overcome the limitations of individual machine learning models and static ensemble strategies, a novel Dynamic Weighted Ensemble (DWE) model is introduced. The model dynamically allocates weights to top-performing algorithms based on their 10-fold cross-validated RMSE, ensuring robust and adaptive prediction performance. Five models-Linear Regression, Ridge Regression, Decision Tree, Random Forest, and Gradient Boosting-were evaluated over ten folds. Based on their average RMSE values, the top three models (Gradient Boosting: 8.25, Ridge Regression: 8.30, Linear Regression: 8.31) were selected. The DWE model, applied on five-fold unseen test data using dynamically computed weights, achieved an average RMSE of 8.07. This reflects a 2.1 % improvement in RMSE and a 2.3 % increase in prediction accuracy over the best individual model. The gains were statistically significant (p < 0.05) based on paired t-test analysis, confirming that DWE offers superior prediction consistency. The proposed DWE model supports real-time optimization in green sand casting, helping reduce defects and improve quality outcomes. It aligns with Industry 4.0 objectives by promoting automated, data-driven decision-making and smart manufacturing practices.•Proposed a novel Dynamic Weighted Ensemble (DWE) model for improved defect prediction in green sand casting.•Achieved a 2.1 % RMSE reduction and 2.3 % accuracy gain over the best individual model with statistical significance (p < 0.05).•Supports Industry 4.0 by enabling real-time, data-driven decision-making in smart manufacturing.