Abstract
This study used machine learning (ML) methods to evaluate the strength and SHapley Additive ExPlanations (SHAP) technique to study the effect of raw materials of cement-based composites (CBCs) incorporating eggshell powder (ESP). Dataset needed for this research was developed from an
experimental study. Two ML techniques were used for modeling, i.e., multilayer perceptron neural network (MLPNN) and extreme gradient boosting (XGB), for the strength evaluation of CBC containing ESP. The ML techniques were validated by examining the difference among actual and estimated strength,
comparison of the coefficient of determination (R2), statistical tests, and k-fold methods. It was noted that the MLPNN prediction model had a satisfactory level of exactness, but the XGB technique forecasted the strength of ESP-based CBCs with a higher level of exactness.
The SHAP evaluation revealed that the most positive impact on the strength was that of cement, whereas fine aggregate had a negative impact. Therefore, it may be concluded that using ESP as a replacement for fine aggregate will result in higher material strength than using it as a replacement
for cement.