A Time Series Approach for Forecasting the Weekly Percentages of Influenza in Canada

Abstract

Influenza is the most common respiratory virus in Canada. This work is an attempt to establish a time series model, based on weekly reports from September 7, 2003 to August 23, 2015 of percentage of flu in Canada (from Canada’s FluWatch). As expected the weekly percentage of flu is seasonal, and our analysis show that a SARIMA (2,1,2)x(0,1,1)52 model is the best fit. Using the proposed model we obtain accurate 10 weeks ahead forecasts. The ability to forecast the rate of flu is important for conducting preventive measures which will lower the incidence of flu. We have conducted also a frequency domain analysis and a cross correlation analysis between the data from Canada’s FluWatch and Google Flu Trends.

Discipline: Statistics

Faculty Mentor: Dr. Cristina Anton