Waste Management Optimization: Detect Organic and Inorganic Waste Types with YOLOv8 on Android

Abstract

Poorly managed waste presents significant environmental and public health challenges in Indonesia, exacerbated by population growth and increasing waste volumes. This study proposes a user-centered, mobile-based waste detection system using YOLOv8 (You Only Look Once version 8) to classify organic and inorganic waste, bridging the gap between model performance and real-world usability. The system is designed for on-device deployment, enabling real-time waste classification under resource-constrained mobile environments. We used a dataset of 8,767 waste images to train the YOLOv8 model in Google Collab. The dataset was split into 70% for training, 20% for testing, and 10% for validation. The model was trained for 25 and 50 epochs, showing improved performance over time. At epoch 50, the model achieved a precision of 0.81, recall of 0.61, mAP50 of 0.69, and mAP50–95 of 0.50, indicating reliable detection under moderate conditions while highlighting challenges in more complex scenarios. Real-time testing under controlled indoor conditions showed detection rates of 83% for inorganic waste and 92% for organic waste. These results reflect system performance in a constrained testing scenario and should not be directly compared with standardized evaluation metrics such as mAP50. The findings demonstrate the system’s potential to improve waste sorting processes and support sustainable waste management. However, challenges remain in improving recall, reducing false negatives, and expanding the dataset. Future work should focus on enhancing detection accuracy, applying advanced data augmentation techniques, and optimizing the application for scalability and real-world deployment.