Predicting Peak Demand Days for Asthma-Related Emergency Hospitalisations: A Machine Learning Approach

Abstract

Predictive analytics in the realm of health has taken on a critical role in disease prevention, particularly concerning prevalent ailments in society. The World Health Organization reports that asthma affected approximately 262 million individuals in 2019, resulting in 455,000 deaths. Each year, asthma-related concerns account for over one million visits to emergency departments, as highlighted by the American College of Asthma, Allergy & Immunology. For developing the model, we are using a configurable algorithm called Predicting Peak Demand (PPD) to anticipate days with elevated asthma-related Emergency Department (ED) visits and hospitalisations in Auckland, New Zealand. By leveraging diverse data sources, the model acts as a valuable planning tool for public health providers, particularly during periods of heightened demand, such as instances of overcrowding witnessed during cold weather or disease outbreaks like COVID-19. The PPD algorithm employed in this model effectively forecasts asthma-related hospitalisations. Meteorological factors, air quality data, and Google trends constitute the sources utilised in building the model. Comparative analysis was conducted using various machine learning algorithms, including ensemble modelling, with a dataset spanning 1,097 continuous days. Techniques like SMOTE and Random Over-Sampling were implemented to address class imbalance to generate synthetic data for the minority class. Experimental evaluation reveals that this model achieves an accuracy of approximately 91.00% in predicting peak demand days for asthma-related hospitalisations. The utilisation of this diverse data-driven model for predicting adverse events like asthma-related overcrowding at a population level can lead to timely interventions, mitigating the socio-economic impact of asthma.