Leveraging Convolutional Neural Networks for Multiclass Waste Classification

Abstract

The impact of population growth on waste production in Indonesia emphasizes the urgent need for effective waste management to mitigate environmental and health risks. Segregating waste into organic and inorganic categories is essential for sustainable management, enabling processes like composting and recycling. Employing convolutional neural networks (CNN) through machine learning presents a promising solution for waste classification. This study utilizes a CNN algorithm to achieve significant accuracy and precision in multi-class waste classification, with particular attention to areas for improvement, such as cardboard classification. Based on the MobileNetV2 architecture and Adam optimizer, the model demonstrates high accuracy and precision, with training and validation accuracy of 95.28% and 89.48%, respectively. High precision and recall values confirm its accurate waste classification. The evaluation of unseen data maintains an accuracy of 86.36%, indicating its generalization ability. However, variations in accuracy among waste classes suggest opportunities for refinement, particularly in cardboard classification.