Optuna ML Framework for SCBA Concrete Strength

Abstract

The cement industry plays a crucial role in global CO2 emissions. As demand for cement continues to rise, innovative solutions are required to mitigate its environmental impact. Over a decade ago, the Paris Agreement (2015) established clear targets for reducing global carbon emissions. In response, the construction industry has adopted strategies such as incorporating agro-waste by-products, including sugarcane bagasse ash (SCBA), as sustainable alternatives to cement and fine aggregate in concrete production. However, variations in material characteristics and pozzolanic reactivity across mix designs make predicting the compressive strength of SCBA concrete using conventional trial-and-error methods challenging. This study explores the use of four Optuna-guided machine learning (ML) models, random forest, support vector regression, extreme gradient boosting (XGBoost), and k-nearest neighbors, to predict the compressive strength of SCBA concrete using a dataset of 844 data points extracted from experimental studies published between 2015 and 2025. Model development and hyperparameter optimization were performed using Optuna, an open-source Python framework, to minimize prediction error and enhance model accuracy. Among the models, XGBoost achieved the highest predictive performance, with a R2 value of 0.966. Interpretability techniques, including feature importance analysis, Shapley additive explanations, and partial dependence plots, revealed the relative influence of individual features on model prediction, identifying the water-to-binder ratio and curing age as the most dominant factors. These findings demonstrate the reliability of ML models for data-driven design of SCBA concrete and highlight their contribution to advancing sustainable construction materials in support of sustainable development goal (SDG) 13 on climate action.