Dimensionality reduction and prediction of soil consolidation coefficient using random forest coupling with Relief algorithm

The consolidation coefficient of soil ( C v) is a crucial parameter used for the design of structures leaned on soft soi. In general, the C v is determined experimentally in the laboratory. However, the experimental tests are time-consuming as well as expensive. Therefore, researchers tried several ways to determine C v via other simple soil parameters. In this study, we developed a hybrid model of Random Forest coupling with a Relief algorithm (RF-RL) to predict the C v of soil. To conduct this study, a database of soil parameters collected from a case study region in Vietnam was used for modeling. The performance of the proposed models was assessed via statistical indicators, namely Coefficient of determination ( R 2), Root Mean Squared Error ( RMSE), and Mean Absolute Error ( MAE). The proposal models were constructed with four sets of soil variables, including 6, 7, 8, and 13 inputs. The results revealed that all models performed well with a high performance ( R 2 > 0.980). Although the RF-RL model with 13 variables has the highest prediction accuracy ( R 2 = 0.9869), the difference compared with other models was negligible (i.e., R 2 = 0.9824, 0.9850, 0.9825 for the cases with 6, 7, 8 inputs, respectively). Thus, it can be concluded that the hybrid model of RF-RL can be employed to predict C v based on the basic soil parameters.