MSW-Net: A hierarchical stacking Model for automated municipal solid waste classification

ABSTRACT

Efficient solid waste management is crucial for urban health and welfare in the midst of fast industrialization and urbanization. In this changing environment, government authorities have a significant role in addressing and reducing the effects of solid waste. While waste separation at the source simplifies processes, manual sorting is a consequence of ignorance in numerous regions, which endangers the health of waste pickers. This study addresses the challenges by introducing the MSW-Net model, a hierarchical stacking model designed for the automated classification of municipal solid waste (MSW). Customized Convolutional Neural Network (custom CNN) and Bayesian-Optimized MobileNet models serve as the base models, with Gradient Boosting employed as the meta-classifier. The MSW-Net model, as proposed, exhibits exceptional performance, attaining 99%, 95%, and 96.43% accuracy rates over training, validation, and testing, respectively. Additionally, the model achieves precision, recall, and F1 scores of 96.42%, 96.43%, and 96.42% during the testing phase. Therefore, the proposed MSW-Net model performs better than the other existing models in sorting the waste. This could also aid the municipal authorities in classifying the waste with minimal human intervention.

Implications: The MSW-Net model, featuring a hierarchical stacking approach with custom CNN and Bayesian-Optimized MobileNet base models, and Gradient Boosting as the meta-classifier, achieves remarkable accuracy in automated municipal solid waste classification. With performance metrics of 99% accuracy in training, 95% in validation, and 96.43% in testing, alongside precision, recall, and F1 scores around 96.42%, the MSW-Net model significantly outperforms existing models. This advancement promises to aid municipal authorities in efficient waste management, reducing reliance on manual sorting and thereby improving the health and safety of waste pickers.

Disclaimer

As a service to authors and researchers we are providing this version of an accepted manuscript (AM). Copyediting, typesetting, and review of the resulting proofs will be undertaken on this manuscript before final publication of the Version of Record (VoR). During production and pre-press, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal relate to these versions also.

Author Contributions

Vaishnavi Jayaraman: Conceptualization, Methodology, Data Curation, Visualization, Investigation, Software and Writing - Original draft preparation.

Arun Raj Lakshminarayanan: Supervision, Validation, Reviewing and Editing.

Availability of data and materials

The dataset generated and/or analyzed during the current study is available from the corresponding author on reasonable request.

Code availability

The code generated during the current study is available from the corresponding author on reasonable request.

Additional information

Funding

This research received no external funding.