Leveraging LSTM and ensemble classifiers for enhanced food waste classification

A long short-term memory (LSTM)-enhanced feature extraction technique was combined with ensemble classifiers (gradient boosting, logistic regression, random forest, and support vector machine (SVM)) to improve waste classification outcomes. The proposed feature extraction approach detects temporal patterns in feature data, improving the robustness of decisions and making this method suitable for waste classification applications. The dataset comprised four food waste classes: packaged items, perishable goods, ready-to-eat, and shelf-stable products. The model was evaluated in terms of accuracy, precision, recall, and F1 score. Gradient boosting and SVM demonstrated the best classification achievement, with precision scores of 0.99, while achieving recall and F1-scores of 0.99. Random forest followed closely with comparable values of 0.96. However, the performance of logistic regression fell short, with 0.88 accuracy, and its identification of perishable items proved particularly difficult. The study shows that ensemble models, particularly gradient boosting and SVM, perform well for waste classification tasks. The results underline the critical role of sophisticated feature-learning methods such as LSTM, which boosts classification performance for efficient waste management technology development.