Treatment-Aware Interpretable Machine Learning Prediction of Mechanical Properties of Banana Fiber Reinforced Concrete

Banana-fiber-reinforced concrete (BFRC) has emerged as a sustainable construction material because of its enhanced tensile performance, crack-bridging capability, and utilization of low-carbon agricultural waste. However, its mechanical properties are strongly influenced by fiber dosage, water – cement ratio, aggregate composition, and fiber–matrix bonding under NaOH treatment. This study developed interpretable machine learning models to predict the compressive strength (CS), split tensile strength (STS), and flexural strength (FS) of BFRC using experimental datasets. The input parameters included cement, fine aggregate, coarse aggregate, water – cement ratio, banana-fiber dosage (BF), and NaOH treatment conditions. Four regression algorithms, linear regression, ridge regression, lasso regression, and elastic net, were implemented with nested cross-validation to improve generalizability and minimize multicollinearity effects. Lasso regression achieved the highest STS prediction accuracy (R2 = 0.9811, RMSE = 0.0073 MPa), while ridge regression performed best for FS (R2 = 0.8668). Elastic net provided the highest CS prediction accuracy (R2 = 0.6671). SHAP and residual analyses revealed that banana fiber dosage and NaOH treatment significantly influenced tensile and flexural properties, whereas the water – cement ratio primarily governed compressive strength. Treated BFRC mixtures also exhibited lower residual variability, indicating improved prediction reliability and enhanced fiber–matrix compatibility.