Predicting concrete compressive strength under thermal exposure using hybrid machine learning and PSO models

The deterioration of concrete at elevated temperatures poses a major challenge to structural safety in fire-exposed and high-thermal architectural environments. This study develops a new machine learning framework that incorporates five hybrid AI–PSO models (GBR-PSO, ANN-PSO, XGB-PSO, RF-PSO, and SVM-PSO). These models are used to simultaneously predict the residual compressive strength, production cost, and embodied CO 2 emissions of thermally affected concrete. A comprehensive experimental database consisting of 207 samples exposed to temperatures up to 1,000 °C was employed for model training and validation. In addition, a practical graphical user interface (GUI) platform was implemented to facilitate real-time engineering prediction and mix optimization. The proposed framework represents an integrated AI-based multi-objective prediction system that supports both mechanical performance evaluation and sustainability assessment of fire-exposed concrete. The results indicate that the GBR-PSO model achieved the highest predictive accuracy for compressive strength, with an R 2 value of 0.992 and an RMSE of 1.890 MPa, outperforming the other models across nine statistical indicators. ANN-PSO and XGB-PSO also demonstrated high reliability in predicting production cost and CO 2 emissions, with R 2 values of approximately 0.95–0.96, enabling informed and sustainable material selection. Sensitivity analysis confirmed that temperature is the dominant factor governing post-fire strength degradation, while optimized constituents such as fine aggregates and fly ash contributed to enhanced residual performance. These findings provide practical value for architectural and structural engineering applications, particularly in high-rise buildings, tunnels, and critical infrastructure. In such systems, ensuring fire resistance, maintaining load-bearing capacity, and reducing environmental impact are essential. By enabling accurate prediction and optimization prior to construction, the proposed models support safer design decisions aligned with contemporary sustainability goals.