Prediction of soil thermal conductivity using individual and ensemble machine learning models

Soil thermal conductivity (λ) is an important parameter in thermal calculation and temperature-field analysis in geotechnical engineering. To accurately predict it, this paper uses individual and ensemble machine learning methods to establish predictive models. The λ measurements were obtained (n = 337) and saturation, dry density, quartz content, sand content and clay content were selected as input parameters for the predictive models. The performance of the prediction model is evaluated by Inspection parameters. The predictive model was k-fold cross-validated and compared with traditional empirical models. The results show that individual and ensemble machine learning models accurately predict λ. The random forest model had the best predictive accuracy, with a correlation coefficient R2 = 0.979, root mean square error (RMSE) = 0.097 Wm−1 K−1 and mean absolute error (MAE) = 0.07 Wm−1 K−1. The MLR model had the worst predictive accuracy. The accuracy of ensemble machine learning models was obviously better than those of individual machine learning models. The accuracy of the proposed predictive model was significantly higher than that of the traditional empirical model. According to the results of performance tests and k-fold cross-validation, the RF and decision tree + bagging models are recommended for predicting λ.