Case Studies in Construction Materials (Jul 2024)
Predicting engineering properties of controlled low-strength material made from waste soil using optimized SVR models
Abstract
In this study, the properties of controlled low strength material (CLSM) made from waste soil were examined, utilizing 53 different mix proportions, and a dataset was subsequently constructed. Models, such as particle swarm optimization (PSO)-support vector regression (SVR), genetic algorithm (GA)-SVR, and grid search (GS)-SVR, were developed to predict both flowability and unconfined compressive strength (UCS). To assess the models’ performance, comparisons between experimental and predicted values, residual distribution histograms, and evaluation criteria were conducted. Results revealed that all three models effectively predicted the flowability and UCS of CLSM made from waste soil. Predicted values closely aligned with experimental values in both training and testing sets. Notably, GA-SVR exhibited greater accuracy when compared to PSO-SVR and GS-SVR. Evaluation criteria, including correlation coefficient (R2), mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) for GA-SVR in assessing flowability were 0.948, 77.899, 8.826, 7.608, and 3.061 %, respectively. Correspondingly, for UCS, the values were 0.934, 0.032, 0.178, 0.115, and 8.956 %, respectively. Furthermore, the histogram depicting the residual distribution of the GA-SVR model closely resembled a normal distribution, characterized by a small mean and standard deviation. Extending the analysis to 45 compression strength data points obtained from public literature, the three models continued to demonstrate effective predictive capabilities, with the GA-SVR model outperforming the others. The evaluation criteria, including R2, MSE, RMSE, MAE, and MAPE in this context were 0.909, 0.011, 0.105, 0.085, and 15.502 %, respectively. Shapley additive explanation (SHAP) values highlighted water as the most crucial input variable in the flowability model. Additionally, water, superplasticizer (SP), and sand positively impacted flowability, whereas soil, fly ash (FA), and cement exerted negative effects. In the UCS model, curing age (CA) emerged as the most important input variable. CA, cement, FA, sand, and soil positively influenced UCS, while water and SP had negative effects. This study is of considerable importance in the accurate prediction and evaluation of CLSM engineering properties, as well as in optimizing mix proportions.
