Advancing biomedical waste classification through a hybrid ensemble of deep Learning, reinforcement Learning, and differential evolution algorithms.

The complex nature of pharmaceutical and biomedical waste poses significant challenges for effective management, particularly in the safe and cost-intensive disposal of infectious materials. This research presents a novel classification model that utilizes a double heterogeneous ensemble integrating deep learning, reinforcement learning, and differential evolution algorithms for waste classification. The model operates through three key phases: image augmentation, ensemble image segmentation, and ensemble convolutional neural network architectures, employing decision fusion techniques that incorporate reinforcement learning and differential evolution. It integrates various image segmentation methods, including U-Net, Mask R-CNN, DeepLab Version 3 Plus, and convolutional neural network architectures such as Inception Version 3, Residual Network 50, Mobile Network Version 2, and Densely Connected Convolutional Network 121.The developed model powers the "Biosorter," a machine specifically designed to differentiate between infectious and non-infectious waste. Comprehensive evaluations conducted on both proprietary and benchmark datasets demonstrate that the proposed BioSorter model significantly outperforms several widely used deep learning architectures-including ResNet50, DenseNet121, MobileNetV2, InceptionV3, and ResNeXt-50. On average, the model achieved classification improvements of 5.35% and 9.05% in accuracy over these methods on the respective datasets. During real-world deployment at a small medical center, the BioSorter achieved 98% sorting accuracy and a 50% increase in processing throughput. Furthermore, a post-deployment usability assessment was conducted using the System Usability Scale (SUS)-a standardized questionnaire commonly used to evaluate perceived ease of use of interactive systems. The BioSorter received a score of 93.5 on the SUS (out of 100), reflecting a high level of user-perceived efficiency and interface simplicity during operational use. This study represents a significant advancement in waste management technology, offering potential to reduce disposal costs and enhance sustainability in healthcare environments.