A machine learning framework for predicting the fire response of concrete-filled steel plate composite walls

Finite element (FE) modelling of concrete-filled steel plate composite (CFSPC) walls under fire is computationally demanding, while experimental data remain limited because of the cost and complexity associated with fire testing. This study presents a machine learning (ML) framework for predicting the fire response of axially loaded CFSPC walls subjected to two-sided ISO 834 standard fire exposure. As part of dataset development, 255 additional FE simulations were generated using the validated FE model and combined with existing data to form an augmented dataset of 490 cases. The predictive framework operates in two stages. First, the fire resistance period (FRP) is predicted using ML algorithms. Then, axial displacement–time curves are predicted in the next stage using a characteristic point-based method and a principal component analysis (PCA)-based method. Support Vector Regression (SVR) achieved the highest accuracy for Stage 1 FRP prediction, with a test R 2 of 0.987, while the PCA-based approach achieved superior accuracy for Stage 2 curve reconstruction, with a mean NRMSE of 0.231. Finally, the trained models were implemented in a web-based prediction application, providing an efficient surrogate tool for rapid prediction of both FRP and deformation response of CFSPC walls under fire within the investigated parameter range.