AI-blockchain synergy for sustainable construction: A ResNet50-based framework for waste classification and traceability

The construction industry consumes a large amount of resources and generates significant waste, making efficient waste management an urgent challenge within circular economy frameworks. This paper presents a novel integrated framework that combines deep learning, blockchain technology, and a custom-built user interface for the classification and traceability of construction materials. Among the models evaluated, EfficientNet-B0 delivered the best performance, achieving a validation accuracy of 82.55 % and a ROC-AUC of 0.968, which demonstrates strong potential for real-world deployment. While ResNet50 serves as a reliable baseline and ViT-B/16 introduces architectural novelty, EfficientNet-B0 offers the most effective balance between accuracy and computational efficiency. Class-level analysis shows excellent performance for well-represented categories such as plastic and metal, but ongoing challenges remain for underrepresented classes, including waste, wires, and wood, due to dataset imbalance and feature overlap. To overcome these limitations, data augmentation and class-balancing strategies are recommended to strengthen classification robustness. The integration of blockchain ensures transparency, immutability, and traceability of classification outputs, while the user interface supports practical use in real-world settings. Overall, the findings demonstrate the feasibility of combining artificial intelligence and blockchain to enable reliable, transparent, and scalable systems for construction waste management.