Thermal performance optimization of cementitious composites incorporating nanoscale carbon-based materials through a regression-driven inverse-design framework

• Inverse design predicts thermal performance of NCM composites. • Ensemble regression achieves the highest model accuracy (R² = 0.99). • Genetic algorithm effectively optimizes mix proportions. • Experimental validation confirms predictive model reliability. A regression-modeling-driven inverse-design framework is proposed to predict and optimize the thermal performance of cementitious composites incorporating nanoscale carbon-based materials (NCMs). An experimental dataset consisting of nine material and processing variables across 172 mix–voltage combinations was used to train five regression models. Among them, the ensemble regression model demonstrated the highest predictive accuracy and was selected as the core forward model. The selected model was integrated with four optimization algorithms—genetic algorithm, Bayesian optimization, particle swarm optimization, and simulated annealing—to identify optimal carbon concentration, nanomaterial length, and supply voltage combinations. Experimental validation across nine mix designs confirmed strong agreement between the predicted and measured temperature variations, with errors ranging between 0.67 and 4.05 °C. The results demonstrate that the proposed data-driven inverse-design workflow provides reliable design-to-target mapping and supports mix-design optimization for NCM-enhanced cementitious composites aimed at controlled heat generation and functional performance.