Authors: Sukanya H N, Assistant Professor, Pavan Kumar T S, Prajwal S Shetty, Pranay Ekunde, Sanjay M
Abstract: Improper waste disposal remains one of the most pressing environmental challenges in both urban and rural settings, contributing to pollution, health hazards, and reduced recycling efficiency. Traditional manual waste segregation is error-prone, labour-intensive, and cannot scale to the volumes of waste generated daily. This paper presents EcoSort, an AI-powered full-stack web application that automates waste classification using a fine-tuned MobileNetV3 Large deep learning model trained via transfer learning. The system classifies waste images into three categories—Recyclable, Non-Recyclable, and Hazardous—achieving approximately 94 % overall accuracy with precision values of 0.95, 0.94, and 0.94 respectively. EcoSort integrates real-time webcam-based detection, a microservices architecture (React/Vite front-end, Node.js/Express back-end, Flask AI service, MongoDB Atlas), Role-Based Access Control (RBAC), JWT authentication, and perceptual hashing (pHash) for duplicate-image detection. A gamification layer comprising reward tiers (Bronze to Platinum), a coupon marketplace, and a community leaderboard motivates responsible waste disposal. Load testing confirmed stable operation under 100 concurrent users with average response times below 3.5 seconds. The platform aligns with UN Sustainable Development Goals SDG 3, SDG 11, SDG 12, and SDG 13, offering a scalable, intelligent pathway toward smarter waste management.
