Efficient Hyperparameter-Tuned Convolutional Neural Network for Waste Classification

Today, increasing consumption habits and industrial activities have made waste management a major environmental issue on a global scale. Traditional waste classification methods cannot provide sufficient efficiency due to high labor costs and human errors. Our primary goal in this study is to design a new custom convolutional neural network (CNN) model that yields the highest accuracy and to find the optimal hyperparameters for this purpose. Using the TrashNet dataset, classification was performed in six classes: cardboard, glass, metal, paper, plastic, plastic and other wastes. We also compared the proposed custom CNN model with the seven pre-trained CNN models in the literature. According to the experimental results, the proposed custom CNN model showed the highest success with 95.03% accuracy and 95% F1 score. The results show that deep learning methods can work with high accuracy in waste classification problems and provide more reliable results compared to traditional methods.