Glass fibre concrete: Experimental investigation and predictive modeling using advanced machine learning with an interactive online interface

This study investigates the mechanical properties of glass fibre concrete (GFC) through experimental and predictive analysis using advanced machine learning (ML) techniques. The experimental work focuses on the mechanical and durability characteristics of GFC. The data obtained in the experimental testing were added to the dataset collected from the literature for the application of machine learning algorithms. The dataset contains 108 compressive strength and 87 split tensile strength data points that evaluated vital factors, including fly ash, cement, aggregates, water, fibre content, superplasticizer, fibre length, fibre diameter, and micro-silica. Optuna, a state-of-the-art hyperparameter optimization library utilizing deep learning, was employed to determine the optimal hyperparameters for each model. The best hyperparameters were selected based on the highest average performance from 5-fold cross-validation. Experimental results showed significant influences of fibre content on GFC mechanical and durability characteristics. The Gradient Tree Boosting Regression (GTBR) model was identified as the optimal model for predicting the compressive and split tensile strength of GFC. The model demonstrated high predictive accuracy for both compressive and split tensile strengths, with R2 values of 0.968 and 0.954, respectively. Shapley Additive exPlanations (SHAP) analysis emphasized the significant impact of fine aggregate, cement, and the amount of glass fibre on both compressive and split tensile strengths, providing valuable insights into the contribution of each feature and enhancing the explainability of the optimum ML model. Finally, a user-friendly online interface was developed, allowing users to predict GFC properties based on the trained GTBR model. This tool, featuring interactive sliders for input variables, ensures precise predictions within the collected data range.