Prediction and validation of geogrid tensile force distribution in back-to-back MSE walls under rail axle load: finite-element and intelligent techniques

In this study, an investigation was conducted to assess the performance of artificial intelligence (AI) and machine learning (ML) methods with distinct characteristics in various problem scenarios. Reinforcement tensile forces play a significant role in the design and performance of retaining walls (RWs). These are crucial for the stability and structural integrity of the retaining walls, preventing wall failure. For this, an attempt was made to predict the reinforcement tensile forces of the back-to-back mechanically stabilized earth (MSE) walls under train loading, which are necessary for upkeeping the transportation infrastructure. Six innovative models were created to counter this challenge that combines AI and ML techniques, i.e., LR, SVM, ANN, ANFIS, ANN-GA, and ANFIS-GA. Consequently, the genetic algorithm (GA) technique was also used to integrate new models, such as GA-ANN and GA-ANFIS. The input data for the models were derived from the parametric study conducted in the finite element analyses. Statistical measures, including root-mean-square-error (RMSE), mean-absolute-error (MAE), and coefficient-of-determination (R²), were analyzed and compared across multiple baseline methods to verify the accuracy of the suggested model. Results show that the proposed model's (ANFIS-GA) accuracy (R²) is 0.9876 and errors (RMSE and MAE) are 0.0191 and 0.0122, respectively. This model outperforms the baseline models in all relevant respects and shall precisely predict the tensile forces of the back-to-back MSE walls.