An interpretable attention-based TabTransformer framework with feature fusion for green architecture classification

In this study, we propose a deep learning-based transformer framework for the automated classification of green vs non-green architecture to analyze environmental effects. The environmental significance of this research work lies in enhancing sustainability by enabling scalable, data-driven assessments of building projects which minimize the reliance on costly and manual certification methods. To address the limitations of existing approaches, we introduce EcoArch- TabFusionNet, a Tab transformer based Fusion Network that fuses grouped architectural, energy, and contextual attributes through a specialized attention-based TabTransformer architecture. The model integrates key feature domains based on Energy, Eco-Tech, Design, and Context while employing dimensionality reduction using Principal Component Analysis (PCA) to minimize redundancy and enhance generalization. The framework is further integrated with explainable AI (XAI) components, including attention heatmaps, LIME, and SHAP, to ensure transparency in feature contributions and decision logic. Proposed model performs late fusion classification across refined feature groups and achieves superior results accuracy of 97.5%, compared to four tabular transformer baselines.