Concrete compressive strength classification using hybrid machine learning models and interactive GUI

Abstract

Abstract Concrete Compressive Strength (CCS) is a critical parameter in structural engineering, influencing durability, safety, and load-bearing capacity. This study explores the classification of CCS using hybrid Machine Learning (ML) techniques and an interactive Graphical User Interface (GUI). Advanced ML algorithms: Random Forest (RF), Adaptive-Boosting (AdaBoost), Extreme-Gradient-Boosting (XGBoost), Light-Gradient Boosting Machine (LightGBM), and Categorical-Boosting (CatBoost) were applied to categorize strength into Low, Normal, and High classes. The dataset, comprising 1298 samples, was split into 80% training and 20% testing for evaluation. Hyperparameter tuning was applied using Bayesian Optimization with fivefold stratified cross-validation, resulting greatly improved the model’s performance. Results showed that LightGBM achieved the highest accuracy, with scores of 0.931 (Low), 0.865 (Normal), and 0.935 (High), and corresponding area under the curve values of 0.967, 0.938, and 0.981. CatBoost also performed well, particularly in Normal and High strength classes, while XGBoost showed slight overfitting in the Normal class. RF and AdaBoost had acceptable performance but struggled with boundary cases. To interpret model predictions, SHapley-Additive-exPlanations (SHAP) analysis was used. Curing duration and cement content were the most influential factors across all strength classes, while water content and superplasticizer played secondary roles. Coarse aggregate became more significant in High-Strength Concrete (HSC). A GUI was developed to allow practitioners to input data and receive real-time strength classifications, bridging the gap between machine learning and practical applications in concrete design.