Estimation of recompression coefficient of soil using a hybrid ANFIS-PSO machine learning model

Recompression coefficient (Cr) is =an essential parameter utilized to predict consolidation settlement of over-consolidated soil. Thus, the main aim of this work was to estimate accurately the Cr, using a hybrid ANFIS-PSO Machine Learning (ML) model that is a hybridization of Adaptive Neuro-Fuzzy Inference System (ANFIS) and Particle Swarm Optimization (PSO). To compare the performance of proposed model, we selected two other benchmark ML models: single ANFIS and Support Vector Machines (SVM). We collected and utilized the data of304 soil samples tested from various construction projects in Vietnam, which included 12 input variables (soil parameters) and one output variable (Cr). Validation indices, namely Correlation Coefficient (R), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE) were utilized for the validation of the model’s performance. Correlation analysis of feature selection results indicated that seven input variables namely clay content, plasticity index, liquidity index, degree of saturation, specific gravity, dry density, and bulk density were of importance and thus selected for prediction of the Cr. Validation results indicated that predictive capability of the hybrid ANFIS-PSO model (R = 0.802) is the best in comparison with other two benchmark models namely SVM (R = 0.727) and single ANFIS (R = 0.734). The findings of our study suggest that the ANFIS-PSO is a powerful ML tool for effectively and quickly prediction of the Cr of soil. This can help save time and reduce costs associated with laboratory experiments for determining this important geotechnical parameter.