THE IMPLEMENTATION OF MACHINE LEARNING IN AN INTELLIGENT WASTE SORTING SYSTEM FOR AUTOMATIC WASTE CLASSIFICATION IN CAMPUS AREAS (CASE STUDY: BANDUNG INSTITUTE OF TECHNOLOGY CAMPUS)

• Main Author: Esteria S. Simatupang, Verawati
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/85576
• Institution: Institut Teknologi Bandung

Waste management is one of the significant environmental issues faced by various countries, including Indonesia. The increase in population and human activities has led to a rise in the volume of waste, especially in high-activity environments such as campuses. The Bandung Institute of Technology (ITB) faces challenges in waste sorting, which is currently done manually, requiring a lot of time and effort. Students, as one of the users of waste bins, often find waste sorting inconvenient and are confused about determining the correct waste category. This frequently leads to mistakes in waste disposal. Additionally, data on waste collected at a particular location over a specific period is needed by the Directorate of Facilities and Infrastructure to map waste characteristics and add waste bin categories as needed. This final project aims to develop an intelligent waste sorting system that can reduce the mixing of inorganic waste, provide waste data information, and automatically and accurately separate inorganic waste in the ITB Campus area using machine learning. The system development was carried out using systems engineering methodology, specifically in the concept development and engineering development stages. To build the machine learning model for the intelligent waste sorting system, several machine learning algorithms for object detection were compared. The chosen machine learning model is YOLOv5n, with a precision value of 0.902, recall value of 0.927, F-score of 0.914, mAP50 of 0.948, pre-process time of 0.3 ms, and inference time of 6.3 ms. The system implementation has been tested based on software components, functional requirements, and machine learning accuracy and performance specifications, with the conclusion that the results were 100% successful. The outcome of this development provides a recommendation for further model training with a larger and more diverse dataset and suggests considering the use of Google's Teachable Machine as an auxiliary tool in the machine learning model development.