Chloride durability and CO₂ footprint in marine-grade concrete via intelligent AI-based multi-objective modeling

Concrete durability in chloride-rich environments is crucial for ensuring the longevity of coastal and marine infrastructure. However, conventional evaluation methods are time-consuming and rarely integrate sustainability considerations. This study develops advanced hybrid machine learning (ML) models to predict chloride resistance, non-steady-state chloride ion migration coefficients (Dnssm), production cost and CO₂ emissions in concrete. Five hybrid ML architectures were established, including CNN-LSTM, XGBoost-PSO, and SVM-based combinations. Among them, the CNN-LSTM achieved the best overall performance with a testing accuracy of R² = 0.97 and low error (RMSE ≈ 1.4), demonstrating strong generalization without overfitting. Sensitivity analysis revealed that the water-to-binder ratio is the most influential parameter affecting chloride diffusivity, followed by test age and coarse aggregate content. Moreover, the CNN-LSTM showed excellent capability in sustainability prediction, accurately estimating CO₂ emissions (R² = 0.97) and providing reliable cost predictions (~84 $/m³). These findings highlight the model’s potential as a decision-support tool that balances durability, sustainability, and cost in marine concrete design. The developed ML framework will be integrated into a user-friendly GUI, enabling transparent prediction and open-access data sharing to promote wider adoption in sustainable infrastructure applications.