Machine learning for soil-geosynthetic interface shear strength analysis

This paper presents geosynthetic interface friction angle prediction by using the Random Forest algorithm. A number of 495 interfaces consisting geomembrane and cohesionless soil and fourteen influencing parameters for each interface are used. In the analysis the Pearson's correlation coefficient is used to measure the linear interdependence between each pair made by input-input and input-output parameters. After the linear analysis, an optimized Random Forest has been initialized to make a prediction of the interface friction angle. Random Forest splits the implemented data into training and testing sets and it is observed only for 3% of the training set and 6% of the testing set the estimation has exceeded ±5° from the actual records. R² measures shows strong coherence between predicted and laboratory study friction angles by resulting R² = 0.93 for training and R² = 0.92 for testing set. Thus, the Random Forest has forecasted interface friction angle adequately.