Predictive Modeling of Saturated Hydraulic Conductivity using Machine Learning Techniques

The hydraulic conductivity of saturated soil is a critical parameter for understanding various engineering challenges related to groundwater. Machine learning techniques offer powerful methods to address complex nonlinear regression problems. This study developed three models, namely a Multilayer Perceptron Neural Network (MPNN), a Support Vector Machine (SVM), and a Tree Boost, to predict field saturated hydraulic conductivity using easily measurable soil properties, such as hydraulic conductivity, clay/silt ratio, soil saturation percentage, d90 of grains, liquid limit, plastic limit, soil pH, hydrocarbon anions, chloride ions, and calcium carbonate content. Soil samples were collected from two locations: the El-Nubaria and Sinai regions, located in the western delta of Egypt. To evaluate the performance of these models, five distinct metrics, namely Mean Squared Error (MSE), Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Scatter Index (SI), and Correlation Coefficient (R), were employed along with a Taylor diagram. Among the models tested, the Tree Boost model demonstrated exceptional accuracy in predicting field-saturated hydraulic conductivity, having a lower SI (0.085) compared to the SVM (0.192) and MPN (0.226) models. Moreover, the Tree Boost model exhibited a higher R value (0.99) than SVM (0.981) and MPN (0.974). The Tree Boost results were compared with those of previous models. The findings highlight the effectiveness of the Tree Boost model and suggest its potential as a reliable tool for estimating field-saturated hydraulic conductivity and generating highly accurate predictions.