An Intelligent Retaining Wall Selection System Based on Stacking Integrated Model

The selection of retaining wall systems is still a challenging work due to the various uncertain factors in deep excavation. Previous studies indicate that machine learning technique is an intelligent auxiliary tool, capable of assisting engineers to select appropriate retaining wall systems at the early phase of a project. However, two potential limitations were identified in previous studies. First of all, the conventional machine learning model requires a large amount of dataset to derive reliable results, but the size of the dataset used in previous models was usually less than 300. Second, the prediction model used in previous studies was single model, but the risk of poor generalization easily emerges for single model, thereby, yielding inaccurate prediction results. Therefore, this paper put forward a stacking integrated model to solve the above problems. Stacking integrated model aggregates the advantages of multiple models. We have used the collected data from the questionnaire investigation to train and validate the presented model. The results showed that in the case of a small dataset size, the accuracy of this model was 87.59%, the recall rate was 65.23%, the precision was 79.16%, and the F‐measure was 73.07%, which was improved compared to the other five models. This model has important guiding significance for the selection of retaining wall systems.