Sorting of Hazardous and Toxic Solid Waste (B3) Using Ensemble Learning Techniques with VGG-16 and DenseNet-121 Architecture

Waste has become an increasingly worrying environmental issue, increasing with population growth and human activity. Indonesia faces a waste emergency with a volume of 35.16 million tonnes by 2022. This research uses Convolutional Neural Networks (CNN) with VGG-16 architecture and DenseNet 121 for classification of seven classes of waste. The dataset from Kaggle consists of 2,900 data with unbalanced distribution. However, promising results were obtained through ensemble modelling combining VGG16 and DenseNet 121. The ensemble approach shows potential in overcoming the weaknesses of a single model and creating a more robust and accurate system. In experiments, VGG achieved 0.76% accuracy with 0.27% loss, while DenseNet achieved 0.79% accuracy with 0.68% loss. The ensemble technique improved the accuracy to 0.83 % with a loss of 0.57%, showing a significant improvement in sewage classification performance over using the architectures separately.