Evaluation of Machine Learning-Based Modeling Techniques for Predicting Hydraulic Conductivity of Diverse Gradation Spectrum Sandy Soils

Hydraulic conductivity (K) directly impacts groundwater flow and contaminant transport within the subsurface, making it indispensable for hydrological studies, environmental assessments, and earth system modeling. This study explores the utilization of machine learning models, specifically Artificial Neural Networks (ANN) and Support Vector Machines (SVM), for predicting the hydraulic conductivity of diverse gradation spectrum sandy soils using three predictor variables as \({d}_{50}\) (median grain size), porosity (n), and dry density ( \({\rho }_{d}\) ). The ANN model, designed to emulate human brain functioning, incorporates a multi-layer perceptron architecture with a sigmoidal activation function, enabling it to effectively capture intricate patterns and non-linear relations within the dataset. Employing the radial basis function (RBF) kernel, the SVM model also exhibits strong predictive capabilities. Key statistical performance indices, including R2 values, Mean Squared Error (MSE), and Root Mean Squared Error (RMSE), demonstrate the accuracy and reliability of both models. However, the ANN model outperforms the SVM model in various aspects, it is important to acknowledge that neither model fully accounts for the entire variability in hydraulic conductivity, considering the multifaceted nature of this property influenced by many soil characteristics. Despite this challenge, the ANN model proves to be a promising tool for predicting the hydraulic conductivity of wide gradation spectrum sandy soils.