Prediction of high-temperature creep in concrete using supervised machine learning algorithms

In this paper, supervised machine learning techniques were employed to develop a prediction model for concrete creep at elevated temperatures. Several algorithms were considered including artificial neural networks (ANN), decision tree regression (DTR), random forest regression, linear regression and Gaussian process regression (GPR) models. A dataset of short-term, basic creep was compiled from experimental data reported in the literature for algorithm training. The selected input variables were time, temperature, 28-day compressive strength, compressive stress, fine and coarse aggregate mass, and steel and PP fiber mass. The data was split into training and verification sets, with 70% used for model training and the remaining 30% for model verification. The results demonstrate that ANN, DTR, RFR and GPR models can accurately reflect concrete creep behavior at elevated temperatures based on statistical indicators. However, further analysis showed that the GPR was the best-performing model, providing highly accurate representations of creep behaviour and demonstrating superior performance to existing empirical equations. Conversely, the DTR was found to not accurately reflect experimental data. Additionally, a Shapley additive explanation analysis was conducted that assessed the significance of each input parameter on model prediction. This research highlights the potential of machine learning techniques to accurately model high -temperature concrete creep behavior and thus represents a powerful tool for engineers and researchers.