A performance assessment on the implementation of machine learning techniques for prediction of cohesion in fiber reinforced sandy soil

A predictive model to determine shear strength and mechanical properties of soil-mix material (soil reinforcement) is required in many geotechnical projects especially when the weight of geomaterial is important for stability or drainage purposes. In this paper, several matching learning (ML) techniques namely Chi-squared Automatic Interaction Detection (CHAID), Classification and Regression Trees (CART), Random Forest (RF), Artificial Neural Network (ANN), Support Vector Machine (SVM), and Generalized Linear Mixed Model (GLMM) were used to predict the effects of reinforcement on cohesion (C) parameter in sandy soil. To establish an appreciate database for prediction purposes, several laboratory tests were planned and conducted on sandy soil mixed with fiber and subsequently, soil properties together with their shear strength parameters were measured. The obtained results from laboratory tests showed that fiber percentage, fiber length, deviator stress and pore water pressure have a significant impact on cohesion values and then, the mentioned parameters were set as inputs. According to the most effective parameters of predictive ML techniques, many models were constructed to predict C of the soil. The modelling results showed that the CHAID model provides the best prediction performance of cohesion in the short term and long term. Coefficient of determination of one and system error of zero for both train and test stages of CHAID have confirmed that this model is a perfect, powerful and applicable ML technique. The design process and model development presented in this study can be considered and used by the other researchers or engineers in resolving their complicated issues.