Prediction of Outpatient Visits for Upper Respiratory Tract Infections by Machine Learning of PM2.5 and PM10 Levels in Taiwan

• Main Authors: Yang, Pei-Hsuan, Hsieh, Mi-Tren, Lin, Gen-Min, Chen, Mei-Juan, Yeh, Chia-Hung, Huang, Zhi-Xiang, Yang, Chieh-Ming
• Format: Conference Proceeding
• Language: English
• Subjects: Air pollution; Diseases; Machine learning; PM10; PM2.5; Senior citizens; Springs; Statistics; upper respiratory infections
• ISSN: 2575-8284
• DOI: 10.1109/ICCE-China.2018.8448613

Particulate Matter (PM) 2.5 and PM10 are referred as a mixture of liquid droplets and solid particles in the air with diameters \leq \mathbf{2.5}\ \mu\mathbf{m} and \leq \mathbf{10}\ \mu\mathbf{m}, respectively. Both PM2.5 and PM10 can deposit on respiratory tract and trigger inflammatory reactions, which makes the respiratory tract predisposed to infections. The study used machine learning on daily PM2.5 and PM10 levels of consecutive 30 days from the open website datasets of Environment Protection Administration between Dec. 2008 and Dec. 2016 to predict the subsequent one-week outpatient visits for upper respiratory tract infections (URI) from the Centers for Disease Control (CDC) in Taiwan between Jan. 2009 and Dec. 2016. The weekly URI cases were classified by tertile as high, moderate, and low volumes. In general, both URI burden and PM levels peak in winter and spring seasons. The testing used the mid-month dataset of each season (Jan., Apr., Jul., and Oct.), and the training used the other months datasets. In the nationwide data analysis, PM2.5 and PM10 levels input to the multilayer perceptron (MLP) can precisely predict the degree of URI number for the elderly (89.05% and 88.32%, respectively) and the overall population (81.75% and 83.21%, respectively). In conclusion, machine learning of PM2.5 and PM10 levels could accurately predict the burden of outpatient visits for URI in Taiwan.