Prediction of Compressive Strength of CSGR Based on Optimized Machine Learning with Sparrow Search Algorithm

Abstract

Abstract Cemented sand, gravel, and rock (CSGR) is an emerging dam material, and accurately predicting its compressive strength is essential for reliable mix-proportion design. Using 342 experimentally measured samples, this study develops an SSA-optimized machine-learning framework in which four base models (BP, SVM, RF, and XGBoost) are systematically tuned via the sparrow search algorithm (SSA). After optimization, the <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" alttext="left bracket mu minus sigma comma mu plus sigma right bracket" display="inline" overflow="scroll"> <mml:mo stretchy="false">[</mml:mo> <mml:mi>μ</mml:mi> <mml:mo>−</mml:mo> <mml:mi>σ</mml:mi> <mml:mo>,</mml:mo> <mml:mi>μ</mml:mi> <mml:mo>+</mml:mo> <mml:mi>σ</mml:mi> <mml:mo stretchy="false">]</mml:mo> </mml:math> uncertainty intervals of all models were markedly reduced, with the smallest <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" alttext="mu plus sigma" display="inline" overflow="scroll"> <mml:mi>μ</mml:mi> <mml:mo>+</mml:mo> <mml:mi>σ</mml:mi> </mml:math> value reaching 0.469, and the training-set MAE, RMSE, MAPE, <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" alttext="normal upper R squared" display="inline" overflow="scroll"> <mml:msup> <mml:mrow> <mml:mi mathvariant="normal">R</mml:mi> </mml:mrow> <mml:mrow> <mml:mn>2</mml:mn> </mml:mrow> </mml:msup> </mml:math> , and R values showed substantial improvements. Among the models, SSA-XGBoost achieved the best performance, yielding the lowest prediction errors ( <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" alttext="upper M upper A upper E equals 0.612" display="inline" overflow="scroll"> <mml:mi>MAE</mml:mi> <mml:mo>=</mml:mo> <mml:mn>0.612</mml:mn> </mml:math> , <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" alttext="upper R upper M upper S upper E equals 0.876" display="inline" overflow="scroll"> <mml:mi>RMSE</mml:mi> <mml:mo>=</mml:mo> <mml:mn>0.876</mml:mn> </mml:math> , <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" alttext="upper M upper A upper P upper E equals 10.65 percent sign" display="inline" overflow="scroll"> <mml:mi>MAPE</mml:mi> <mml:mo>=</mml:mo> <mml:mn>10.65</mml:mn> <mml:mo>%</mml:mo> </mml:math> ) and the highest goodness-of-fit ( <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" alttext="normal upper R equals 0.983" display="inline" overflow="scroll"> <mml:mi mathvariant="normal">R</mml:mi> <mml:mo>=</mml:mo> <mml:mn>0.983</mml:mn> </mml:math> , <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" alttext="upper R squared equals 0.964" display="inline" overflow="scroll"> <mml:msup> <mml:mi>R</mml:mi> <mml:mn>2</mml:mn> </mml:msup> <mml:mo>=</mml:mo> <mml:mn>0.964</mml:mn> </mml:math> ). SHAP-based global and local interpretability analysis further identified cement content and curing age as the most influential factors governing CSGR strength. In addition, a graphical user interface (GUI) prediction platform was developed based on the SSA-XGBoost model, and experimental validation indicated that the prediction deviations remained below 10%. Overall, this study provides an accurate, interpretable, and practical tool for CSGR strength prediction and offers technical support for mix-proportion optimization in CSGR dam construction.