Evaluation of Machine Learning Algorithms for Predicting Compressive Strength of Geopolymer Concrete at High Temperatures

In recent years, fly ash and slag, both industrial by-products, have been integral to sustainable geopolymer concrete, but their fire resistance lacks clarity. The complex interaction between its composition and elevated temperatures has a significant impact on its resulting mechanical properties, thereby suggesting the need for precise predictive models. This significant knowledge gap can be bridged by using advanced machine learning methods to accurately forecast the compressive strength of sustainable geopolymer concrete under high-temperature conditions. In this study, an experimental dataset consisting of over 200 data points is collected from literature. Several machine learning techniques, encompassing linear regression, decision trees, support vector machine, Gaussian process regression, ensemble methods, neural networks, and kernel-based algorithms, are employed to forecast the compressive strength. Statistical and performance metrics for all machine learning models are tabulated to examine their efficiency. It is observed that the trilayered neural network model outperformed other algorithms, achieving an R² value of 0.95 and MSE of 37.55. The study provides structural designers with a reliable computational tool for assessing the strength of green concrete exposed to elevated temperatures.