A User-Friendly Machine Learning Framework for Predicting Brunauer–Emmett–Teller Surface Area of Porous Materials

The Brunauer–Emmett–Teller (BET) surface area is an essential indicator of the adsorption capacity and suitability of materials for CO 2 capture. However, accurately predicting BET is challenging due to the nonlinear and interconnected effects of synthesis parameters such as precursor type. BET experiments are time-consuming, costly, and limited in their ability to capture complex parameter interactions. To overcome these challenges, data-driven approaches have emerged as promising alternatives to experimental research. In the literature, some studies used temperature, ratio, and contact time as inputs to predict BET. In addition, they used only one method, which is response surface methodology, without the model’s interpretability and had less accuracy with an R of -0.46. To address these limitations, this study proposes a comprehensive, data-driven framework that employs 30 machine learning techniques to predict BET. This study considers temperature, ratio, KOH activator, precursor type, heating rate, and activation time as predictors of BET. The performance of these models is assessed through different methods, such as trend analysis and statistical error analysis. Among all models tested, the LMB-based artificial neural network demonstrates the highest predictive accuracy with an R of 0.92. The SHAP analysis was used to demonstrate the interpretability of the proposed LMB ANN model. The SHAP study proves that temperature is the most influential factor affecting BET. The precursor concentration and KOH activator content are also important in determining the BET. Time has the lowest contribution among the parameters. Ultimately, this study delivers an easy-to-use predictive tool enabling rapid and precise BET estimation, supporting the optimization and design of high-performance porous materials for CO 2 capture applications.