<![CDATA[The rapid growth of non-organic waste has created a demand for smart automated classification systems to improve waste management. Although deep learning has been widely used, differences in model designs, datasets, and evaluation methods in earlier research have led to inconsistent conclusions about real-world effectiveness. This review analyzes 40 peer-reviewed journals focused on deep learning for classifying non-organic waste. It examines model types, dataset usage, and evaluation metrics to identify research trends and gaps. Results show that Convolutional Neural Networks (CNNs), especially deep architectures like Residual and Efficient Networks, achieve higher accuracy on standard benchmarks. Yet, these models often perform less well with actual waste data, revealing limitations in generalization. Accuracy remains the main metric, while metrics like precision, recall, and F1-score are less consistently reported. Findings highlight that classification success depends more on dataset realism and evaluation design than on model complexity alone. This study offers a structured synthesis to identify research gaps and guide the development of more reliable, practical waste classification systems. From an informatics point of view, it provides a framework for performance evaluation that supports evidence-based model selection and the creation of robust, deployable intelligent classifiers.]]>
Copyrights © 2026
