An efficient long short-term memory-based model for prediction of the load-displacement curve of concrete-filled double-skin steel tubular columns

Concrete-filled double-skin steel tubular (CFDST) columns have been widely used in buildings and infrastructures. Therefore, the compressive behavior of CFDST columns is a subject that should be further investigated. In this paper, an efficient procedure using a Long Short-Term Memory (LSTM)-based model is proposed to predict complete axial load-displacement curves of CFDST columns under compression for the first time. Hybrid databases, including experimental data collected from the literature and numerical data generated from finite element analyses, are utilized to establish the proposed model. To augment the training data, the DoppelGANger (DGAN) algorithm is used for generating additional dummy data, which is added to the training set. After that, the Optuna framework is used to optimize the hyperparameters of the proposed model. The accuracy of the proposed model is demonstrated by comparing its predictive results with those obtained from tests and finite element results. In addition, the importance and contribution of each input variable in the proposed model are evaluated by using SHapley Additive exPlanations (SHAP) method. Finally, a cloud-based graphical user interface (GUI) is developed in the Hugging Face platform to predict and plot the complete axial load-displacement curve of CFDST columns. This GUI is a convenient tool for making predictions without requesting machine learning knowledge by the users.