Comparative Analysis and Evaluation of Deep Learning Models for Efficient Waste Classification

Globally, an estimated 2 billion tons of waste are generated annually, yet only 9 percent of plastic waste is successfully recycled, largely due to improper segregation and misclassification. Our study employs five deep learning techniques i.e., VGG19, InceptionV3, DenseNet121, ResNet50, and EfficientNetB0 to classify waste images as recyclable or non-recyclable. The models are trained and evaluated on three different datasets with separate train and test data to ensure robustness and generalizability. Our study aims to enhance recycling efficiency, reduce contamination in recycling streams, and raise public awareness about proper waste disposal practices. Future advancements include expanding the dataset, developing a mobile app for real-time classification, integrating the model into smart waste bins, and collaborating with local governments to improve community recycling efforts. By leveraging cutting-edge technology, our study aspires to promote a circular economy and foster responsible waste management practices, ultimately supporting broader environmental sustainability goals. Among all the models tested, DenseNet121 emerged as the top performer, achieving a training accuracy of 99.80 and a validation accuracy of 87.36.