Machine Learning Framework to Manage Biomedical Waste Response to Epidemics

Abstract

The COVID-19 pandemic has led to increased hospital admissions and a marked increase in medical waste due to the high consumption of Diagnostic kits, Personal protective equipment, and Medications. This global pandemic has created a demand for innovative solutions to handle biomedical waste-related issues in pandemics, especially when traditional waste management practices are insufficient. The current system for handling waste segregation, correct disposal methods monitoring, and biomedical waste transportation is not enough, and many of its elements are often neglected, leading to unnecessary health and environmental hazards. In light of these challenges, this study introduces a machine-learning framework for managing biomedical waste during epidemics to meet the needs of healthcare facilities and hospitals. This research proposes a machine learning-based framework to address these issues to improve Waste Segregation and Classification, Waste Generation Forecasting, Waste Collection and Transportation Optimization, and Disposal Monitoring and Anomaly Detection. The framework utilizes a Random Forest regression model to predict biomedical waste generation and optimize waste collection and scheduling. Several biomedical waste generation prediction models have been developed in this study to compare actual data collected during the COVID-19 epidemic and hypothetical data generated through a simulated SEIR (Susceptible, Exposed, Infective, Recovered) model. The model based on Random Forests achieved excellent performance in both settings, with 93% accuracy on the actual pandemic training data and 97% on the SEIR model training data. However, the study concluded that real-world data provides more practical and reliable predictions for total biomedical waste volumes. The model was also able to predict the weekly generation of waste and, based on that, calculate the number of trucks necessary for transportation, thus suggesting scheduling and logistic improvements.
Furthermore, this study proposes future advancements by incorporating Convolutional Neural Networks (CNNs) for the automation of waste classification; if implemented, this automated system would assist staff in real-time with 90% accuracy based on a validation set of synthetic data, ensuring that waste is correctly sorted into appropriate containers. Moreover, an isolated forest model was proposed in this study to detect waste disposal violations. This allows waste handlers to take corrective actions before violations become a severe health and environmental hazard. While these methods were not deeply analyzed in this study, their integration into the framework would improve the biomedical waste management framework. With all four framework stages in place, an integrated machine-learning framework could be established and applied to optimize biomedical waste management systems during pandemics. The findings indicate that even if not fully implemented, machine learning can still improve biomedical waste management to an extent where healthcare facilities can take necessary measures for the management of pandemic-related biomedical waste generation, Stocking the required Personal Protective Equipment to ensure staff safety, and strategically schedule waste collection and transportation. The framework would offer a scalable, proactive solution to support healthcare facilities and waste management systems during future health crises.

Date of Award	25 Nov 2024
Original language	American English
Supervisor	Saed Amer (Supervisor)