Leveraging machine learning for prediction and optimization of texture properties of sustainable activated carbon derived from waste materials.

Abstract

The increasing demand for sustainable waste management has driven innovation in the production of activated carbon (AC) from waste. AC's textural properties, including its surface area (SA), total pore volume (TPV), and micropore volume (MPV), are critical for applications such as gas purification and wastewater treatment. However, the traditional assessment methods are expensive and complex. This study employed machine learning (ML) models to predict AC's properties and optimize its production process. Random Forest (RF), Decision Tree (DT), Gradient Boosting Regressor (GBR), support vector machines (SVM), and Artificial Neural Networks (ANN) were applied along with key input parameters, including raw material type, particle size, and activation conditions. A genetic algorithm (GA) integrated with the GBR model optimizes the synthesis process. The ML models, particularly RF and GBR, accurately predicted SA with R² values exceeding 0.96. In contrast, the linear regression models were inadequate, with R² values below 0.6, emphasizing the non-linear relationship between the inputs and outputs. Sensitivity analysis showed that the activation temperature, ratio of the activating agent to carbon, and particle size significantly affected the AC properties. Optimal properties were achieved under activation temperatures between 800 and 900 °C and activating-agent to the carbon ratio 3.8. This approach provides a scalable solution for enhancing AC production sustainability, while addressing critical waste management challenges.