| dc.description.abstract | Waste has become one of the increasingly concerning environmental issues, escalating alongside population growth and human activities. According to data obtained from the Integrated Environmental and Forestry Information System (SIPSN) of the Ministry of Environment and Forestry, Indonesia is currently facing a waste emergency, with a waste generation volume of 35.16 million tons in 2022. Improperly managed waste can have negative impacts on the environment and human health. In this study, Convolutional Neural Networks (CNN) and two CNN architectures, namely VGG-16 and DenseNet 121, were employed. Seven classes of data were used in this research, utilizing Kaggle as the dataset source. The data exhibited an imbalance among the seven classes, with four classes comprising 500 samples each and three classes comprising 300 samples each, totaling 2,900 data points. Despite the imbalanced data, promising results were obtained through ensemble modeling, combining the VGG16 and DenseNet 121 architectures. The utilization of ensemble learning with DenseNet 121 and VGG16 not only significantly improved waste classification accuracy but also demonstrated superiority in performance consistency compared to using each architecture separately, as reported in previous studies. This underscores the potential of ensemble approaches in addressing individual weaknesses of single models and creating a more robust and accurate system. In the experiment results of VGG and DenseNet architectures, accuracy values were observed, where VGG achieved an accuracy score of 0.76%, while DenseNet achieved a better accuracy score of 0.79%, with prediction error rates (loss) for each model below 70%, VGG at 0.27% and DenseNet at 0.68%. To enhance the accuracy of each architecture, ensemble techniques were utilized in this study, yielding favorable results of 0.83% accuracy and a prediction error (loss) rate of 0.57%. | en_US |
