Prediction of uplift resistance of circular anchors in anisotropic clays using MLR, ANN, and MARS

Abstract

A new equation predicting the uplift resistance of circular anchors in anisotropic and heterogeneous clays is proposed in this study using machine learning techniques, i.e., MLR, ANN, and MARS. The uplift resistance is determined through the uplift resistance factor Fc which is set to be the normalized output parameter and functioned by three dimensionless parameters of the embedment ratio, the strength ratio, and the anisotropic ratio. The numerical results obtained from the previously published scientific paper have been used as the training dataset for machine learning models. The potential of machine learning models is examined via traditional statistical indices. Furthermore, the forecast ability and reliability of the established models are also assessed through ranking criteria and validated with external data. As shown in the obtained results, MARS demonstrates its excellent veracity in predicting the uplift resistance of circular anchors in anisotropic and heterogeneous clays. Based on the optimal MARS model, the sensitivity analysis is implemented to detect the impact of each dimensionless input parameter on the uplift resistance. The combination of sensitivity analysis results and novel empirical equations can be an efficient tool and useful theory guideline for engineering practitioners.