Abstract
The compressive strength prediction of Basalt Fiber Reinforced Concrete (BFRC) is a complex task influenced by numerous factors. For traditional empirical formula methods, difficulties such as the limited number of experiments and the time-intensive nature of the process may hinder the accurate prediction of the compressive strength. The compressive strength of BFRC was predicted using a machine learning approach, considering various combinations of characteristics, and the predictive ability of the model was verified using experimental data. A database of 309 sets of BFRC compressive strength data was assembled, and eight sets of BFRC compressive strength experimental data were experimentally obtained. After the hyperparameter optimization four machine learning models were employed to investigate their performances. Combined with the SHAP algorithm, an interpretive analysis of the input parameters of the XGBoost model was specifically conducted, and the predictive ability and interpretability of the model were verified using experimental data. The research results indicated that the XGBoost model surpasses the other three machine learning models in terms of prediction accuracy, achieving an R2 value of 0.9431, RMSE of 3.2325, and MAE of 2.3355. The SHAP analysis revealed that among the basalt fiber (BF) parameters, the volume content of BF had the most significant impact on the XGBoost model output. Furthermore, the optimal range of volume content is 0.1%, the optimal range of diameter is 15–20 µm, and the optimal range of length is 8–15 mm.