Implementation Of Machine Learning To Identify Types Of Waste Using CNN Algorithm

This research aims to develop a waste-type classification model using the Convolutional Neural Network (CNN) method to increase efficiency and accuracy in waste management. The research method involves collecting image waste data from open sources consisting of 2848 files, which are divided into six subclasses: glass, cardboard, paper, metal, organic, and plastic. This data is then processed through preprocessing stages such as cropping and applying a Gaussian filter to remove noise and improve image quality. The training model was performed for 10 epochs with learning rate parameters 0.0001 and epsilon 0.00000001. The model training results achieved the highest accuracy of 95.64% and the smallest loss value of 0.3806 at the 57th epoch. Testing with test data showed strong performance on all evaluation metrics, with an overall accuracy of 95% and an average precision, recall, and F1 score of 95%, indicating that this model can be used for garbage classification. The originality of this research lies in the use of CNN in classifying types of waste based on diverse image data and through an augmentation process to enrich the dataset without collecting new data. This research confirms that increasing the number of epochs does not always produce an optimal model. The loss value parameter is proven to be more effective in determining the most optimal model compared to just relying on training accuracy or the highest number of epochs. These results contribute significantly to technology-based waste management methods, which can be activated in applications such as EcoSortBin to support environmental education and awareness among the public. This research also opens up opportunities for further development in applying deep learning to various other environmental applications.