Hydraulic conductivity of soil reveals its influencing role in the studies related to management of surface and subsurface flow, e.g. irrigation and drainage projects, and solute mass transport models. Direct measurements of hydraulic conductivity have many difficulties due to spatial variation of the property in the field. Pertaining to this problem, in this study, estimation models have been developed using machine learning methods (M5 tree model and random forest model) in an attempt to estimate the accurate values of unsaturated hydraulic conductivity related to basic soil properties (clay, silt and sand content, bulk density and moisture content). Data set was collected from the experimental measurements of cumulative infiltration using mini disc infiltrometer at the study area (Kurukshetra, India). A multivariate nonlinear regression (MNLR) relationship was derived, and the performance of this model was compared with the machine learning-based models. The evaluation of the results, based on statistical criteria (R², RMSE, MAE), suggested that random forest regression model is superior in accurate estimations of the unsaturated hydraulic conductivity of field data relative to M5 model tree and MNLR.

In this investigation, the potential of M5P, Random Tree (RT), Reduced Error Pruning Tree (REP Tree), Random Forest (RF), and Support Vector Regression (SVR) techniques have been evaluated and compared with the multiple linear regression-based model (MLR) to be used for prediction of the compressive strength of bacterial concrete. For this purpose, 128 experimental observations have been collected. The total data set has been divided into two segments such as training (87 observations) and testing (41 observations). The process of data set separation was arbitrary. Cement, Aggregate, Sand, Water to Cement Ratio, Curing time, Percentage of Bacteria, and type of sand were the input variables, whereas the compressive strength of bacterial concrete has been considered as the final target. Seven performance evaluation indices such as Correlation Coefficient (CC), Coefficient of determination (R²), Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Bias, Nash-Sutcliffe Efficiency (NSE), and Scatter Index (SI) have been used to evaluate the performance of the developed models. Outcomes of performance evaluation indices recommend that the Polynomial kernel function based SVR model works better than other developed models with CC values as 0.9919, 0.9901, R² values as 0.9839, 0.9803, NSE values as 0.9832, 0.9800, and lower values of RMSE are 1.5680, 1.9384, MAE is 0.7854, 1.5155, Bias are 0.2353, 0.1350 and SI are 0.0347, 0.0414 for training and testing stages, respectively. The sensitivity investigation shows that the curing time (T) is the vital input variable affecting the prediction of the compressive strength of bacterial concrete, using this data set.