Development of a Machine Learning-Driven Automated Waste Sorting System

Abstract

The rapid growth of municipal solid waste presents environmental and health challenges, while manual sorting remains inefficient and hazardous. This study develops a prototype machine learning–driven automated waste sorting system designed for low‑cost deployment. The system integrates an ESP32‑CAM module for image capture, a MobileNetV2 convolutional neural network (CNN) transfer learning technique for classification, and a servo‑actuated mechanical sorting mechanism. Waste items are classified into biodegradable and non‑biodegradable categories, with system control managed through an ESP32 microcontroller and real‑time communication via WebSocket. The model was trained on 90 live images with an additional 2356 images augmented from an online database. The system achieved an accuracy of 93.57% on the test set. While this result confirms the system's feasibility, its generalizability is currently restricted by lighting sensitivity, a limited dataset, and an inability to classify wet waste. The prototype highlights the potential of combining embedded AI and mechatronics for sustainable waste management, supporting circular economy goals. Future work should expand datasets aimed at mitigating overfitting and addressing high intra-class variance, incorporate synthetic data generation to enhance model robustness, and explore edge inference for reduced latency.