Exploring machine learning models to predict the unfrozen water content in copper-contaminated clays

Abstract

The article provides new insights into predicting unfrozen water content(u_nf) in clays contaminated with copper. The objectives of this study included creating machine learning prediction models based on Gaussian Process Regression (GPR), Support Vector Machine (SVM), and Random Forest (RF) algorithms. These models were developed using seventeen soil physicochemical parameters. A total of 575 experimental observations of unfrozen water content, determined by the DSC method over a temperature range of −23 °C to −1 °C, were analyzed. The findings suggest that the unfrozen water content in copper-contaminated clays can be most accurately predicted using the Random Forest model, which achieved a high correlation coefficient (R = 0.962). This model demonstrated greater effectiveness than existing empirical models in estimating unfrozen water content in these soils. Further research should focus on exploring alternative machine learning techniques to improve predictions of unfrozen water content.