Sustainability-driven tiny deep learning empowering green edge intelligence for smart environments

Rapid urbanization and the growing emphasis on sustainability demand intelligent systems that minimize energy consumption and environmental impact. This study introduces a sustainability-driven Tiny Deep Learning (TDL) framework that empowers green edge intelligence for smart environments through energy-efficient and carbon-aware computation. The proposed approach emphasizes ultra-lightweight model design, integrating quantization and pruning-based compression to achieve sub-kilobyte memory footprints while maintaining high predictive performance for real-time occupancy detection. Comprehensive benchmarking against classical machine learning and TDL baselines under identical deployment conditions demonstrates the framework’s superior balance between accuracy, latency, and energy efficiency. In addition to predictive metrics, explicit evaluation of energy per inference and equivalent CO₂ emissions establishe a transparent sustainability profile, an often-overlooked dimension in prior edge AI research. Real-world simulations within a Building Management System validate significant reductions in daily energy usage and carbon emissions, confirming the framework’s contribution to operational sustainability. Deployment on low power microcontrollers, including Arduino Nano 33 BLE Sense and Raspberry Pi Pico, further affirms its feasibility for large-scale IoT integration. Beyond conventional performance boundaries, this work positions TDL as a scalable, eco-conscious edge AI paradigm aligned with the United Nations Sustainable Development Goals (SDGs) 7, 11, and 13, fostering the next generation of sustainable and intelligent urban infrastructures.