<![CDATA[The increasing volume of waste worldwide has led to significant challenges related to pollution, waste management, and recycling. These issues require innovative solutions to enhance the waste management ecosystem, such as the implementation of Smart Waste Management, which leverages information technology and artificial intelligence. This study aims to implement the Xception Convolutional Neural Network (CNN) model for waste classification, explore various data augmentation techniques, and identify optimal model configurations for this task. The research methodology consists of several stages, including data preparation, model building and training, model adaptation for classification tasks, model evaluation, iterative experimentation, and saving and reloading the trained model. The dataset used in this study is the TrashNet dataset obtained from Kaggle, consisting of 2,527 images across several classes: cardboard, glass, metal, paper, plastic, and trash. Based on the optimization process, the selected hyperparameters include a batch size of 32, 64 convolutional filters, the Adam optimizer (learning rate = 0.0001), and a dropout rate of 0.25. After training for 100 epochs, the model achieved a training accuracy of 99% with a loss of 0.7%, and a validation accuracy of 87% with a validation loss of 52%. Evaluation on the test dataset yielded an accuracy of 76%, precision of 79%, recall of 75%, and an F1-score of 75%. The application of data augmentation techniques—such as scaling, translation, and color space transformation—resulted in performance improvements, increasing accuracy by 13%, precision by 11%, recall by 13%, and F1-score by 12%. This study contributes by implementing the Xception model on the TrashNet dataset for waste classification and proposing several data augmentation methods that provide empirical evidence to support or challenge existing approaches. The findings offer practical insights for the development of Smart Waste Management systems, enrich the literature through experimental results, and provide a comparative analysis of data augmentation techniques suitable for the TrashNet dataset.]]>
Copyrights © 2026
