AI Models Achieve 91% Accuracy in Waste Image Classification for Circular Economy

Global: AI Models Achieve 91% Accuracy in Waste Image Classification for Circular Economy

Researchers Julius Sechang Mboli and Omolara Aderonke Ogungbemi presented a study on Jan. 30, 2026 that evaluates a range of machine‑learning and deep‑learning techniques for binary classification of waste images. The work, later featured at the 2025 IEEE International Smart Cities Conference in Patras, Greece, aims to support circular‑economy initiatives and urban sustainability by automating waste sorting.

Dataset and Preprocessing

The authors assembled a collection of 25,077 waste photographs, which were randomly divided into an 80% training set and a 20% test set. All images were augmented and resized to 150 × 150 pixels before model ingestion, providing a uniform input size for both traditional and deep‑learning pipelines.

Traditional Machine Learning Approaches

Three conventional classifiers—Random Forest, Support Vector Machine, and AdaBoost—were trained on the raw pixel data and on features reduced via Principal Component Analysis (PCA). The analysis reported that PCA offered negligible performance gains for these models, suggesting limited utility of dimensionality reduction under the given conditions.

Deep Learning and Transfer Learning

The study also examined custom convolutional neural networks alongside five pre‑trained architectures: VGG16, ResNet50, DenseNet121, EfficientNetB0, and InceptionV3. Transfer‑learning models were fine‑tuned on the waste dataset, leveraging knowledge from large‑scale image corpora to improve classification under limited‑data scenarios.

Performance Outcomes

DenseNet121 emerged as the top performer, achieving 91% accuracy and a ROC‑AUC of 0.98 on the test split. This result outperformed the best traditional classifier by 20 percentage points, highlighting the advantage of deep‑learning methods for visual waste identification.

Integration into Decision Support Systems

The authors outline how the high‑performing models can be embedded within a real‑time, data‑driven decision‑support platform for automated waste sorting. Such integration is intended to streamline material recovery processes and reduce reliance on manual segregation.

Environmental and Sustainability Implications

By improving sorting efficiency, the proposed system could lower landfill usage and diminish lifecycle environmental impacts associated with waste management, thereby contributing to broader circular‑economy goals.

Future Directions

The paper suggests expanding the image repository, exploring multimodal sensor data, and assessing deployment scalability in diverse urban settings as avenues for further research.

This report is based on information from arXiv, licensed under Academic Preprint / Open Access. Based on the abstract of the research paper. Full text available via ArXiv.