Next-generation cutting-edge hybrid AI frameworks for predicting rheological properties and CO₂ emissions in alkali-activated concrete

The rheological properties of alkali-activated concrete (AAC) are critical for its performance in modern construction applications, where stability, pumpability, and resistance to washout are essential. Given AAC's constituent interactions' complexity and nonlinear nature, conventional modeling methods often fail to predict its rheology accurately. To address this challenge, this study presents a next-generation AI-based predictive framework utilizing three hybrid machine learning techniques: adaptive neuro-fuzzy inference system with genetic algorithm (ANFIS-GA), convolutional neural networks with long short-term memory (CNN-LSTM), and multi-objective optimization (MOO). These models were trained on experimental data to estimate the key rheological parameters and further validated using statistical metrics such as MSE, MAE, RMSE, MARE, precision, recall and F1 score. The results demonstrated that all models achieved high prediction accuracy (R² > 0.9), with ANFIS-GA outperforming the others across all parameters. Sensitivity analysis revealed that precursor content strongly influenced output behavior, particularly within the 400-450 range. Furthermore, the models were extended to predict CO₂ emissions, with ANFIS-GA delivering the most accurate results, ranging between 22.51 and 68.27 kg/m³ , closely matching the experimental values. A Taylor diagram comparison showed ANFIS-GA’s superior performance over CNN-LSTM by 2.41 % (SYS), 1.48 % (DYS), and 1.71 % (PV). Finally, a Python-based graphical user interface (GUI) was developed, making the predictive tools accessible to users without prior AI knowledge. This study offers a new, practical, and sustainable approach for optimizing AAC formulations, contributing significantly to intelligent material design and next-generation construction technologies.