<![CDATA[This study investigates the comparative performance of ten machine learning models—Linear Regression, SVM, Neural Network, Decision Tree, Random Forest, Gradient Boosting, AdaBoost, XGBoost, LightGBM, and CatBoost—in predicting concrete compressive strength. The research emphasizes practical applications in construction, where accurate predictions can improve material design and structural reliability. Through detailed evaluation using MAE, RMSE, and R² metrics, CatBoost and Linear Regression emerged as top-performing models. A rigorous hyperparameter tuning process, employing grid search, significantly enhanced models like SVM and Neural Network, increasing their R² by over 80%. However, tuning occasionally led to reduced performance due to overfitting or unsuitable parameter selection. Outlier analysis using the Z-score method revealed nuanced effects across models: while SVM and Decision Tree benefited from outlier removal, models like Neural Network and CatBoost experienced performance degradation, indicating their reliance on diverse data patterns. These findings underscore the importance of tailored tuning and outlier handling strategies. Future work will incorporate advanced optimization techniques (e.g., Bayesian optimization) and robust cross-validation to further improve model generalization and stability.]]>
Copyrights © 2025
