Abstract:
Increased population, faster urbanization, and industrialization are the main reasons for ecological contamination. The amount of waste products has increased to the point that they pose a threat to both human health and the environment. In this fast-paced reality, it has become an absolute necessity to collect and recycle waste. By encouraging waste management, classifying and discharging wastes contribute to resource preservation, pollution reduction, and energy conservation. Waste items are classified according to the substance and composition of the trash and whether it can be recycled or disposed safely. Till now, the majority of wastes have been classified manually. The best way to classify images effectively is by using Machine Learning (ML). A significant amount of attention has been provided to hyperparameters in ML and hyperparameter tuning is now recognized as a crucial stage in ML process. Model selection through hyper-parameter selections is still a complex and heavily intuition-driven process, even though these models learn their parameters using data-driven methods. The primary objective of this work is to investigate how our selected models' performance for waste categorization is affected by hyperparameter tuning. The goal is to determine the performance improvements attained by using methodical and effective hyperparameter tuning techniques. Of all the models evaluated, our tuned DenseNet169 (D5) stood out as the best performer, with an accuracy of 99.99%, precision of 98%, and recall of 98%. Additionally, this model demonstrated a low loss value of 0.0012 and a high validation accuracy of 97.8%, Our findings will reduce the segregation cost and speed up the process of waste management. This will also reduce environment pollution
