Investigation of Input Number Effect on Performance Prediction of Soil Friction Angle Using Random Forest

Abstract

AbstractThe friction angle is one of the most important parameters for analyzing geotechnical properties. The friction angle of soil depends on numerous factors including liquid limit (LL), plasticity index (PI), deviation of the Casagrande ∆PI, clay fraction (CF). Determining the friction angle of soil is a challenge for geotechnical engineers. In this paper, Random Forest (RF) model is used to predict the friction angle. To develop the model, 131 experimental data were collected from the literature. The collected data is randomly classified into 2 groups for training and testing process. Performance evaluation of the RF models was carried and compared on training dataset (70% data) and testing data set (30% to remaining data) by criteria of coefficient of correlation (R2), root mean square error (RMSE), mean absolute error (MAE). The RF model using 4 inputs gave the best performance. The performance values show that the RF model can accurately predict the friction angle of soil.KeywordsFriction angleMachine learningRandom forestSoilInput number