This README only offers a brief overview of the assignment. Further analysis containing forecasting graphs and literature review can be found in "Disease Forecasting Analysis" file of this repository.
For this assignment assignment, we were given environmental data collected by US Federal Government Agencies. The data included environmental variables surrounding changes in temperature, precipitation, and vegetation. With this data, we were tasked with predicting the number of Dengue fever each week for two cities— specifically for the cities of San Juan, Puerto Rico and Iquitos, Peru. Exploring this problem provides incredibly valuable knowledge and offers immense benefits to communities as a whole. In particular, having accurate predictions of disease cases would result in lowering transmission rates and reducing the amount of fatalities. Knowing the number of cases can aid with controlling outbreaks and reducing the transmission of Dengue fever. This knowledge inherently leads to taking better preventive action to mitigate the risk of being infected. In turn, this can increase the spread of awareness which cotntributes to early detection and earlier treatment— heavily improving the chances of survival. Preparing for the incoming number of cases creates more research incentive as well. This is especially important due to the amount of time, money, and effort research requires. Better research elicits more effective treatment and grants doctors a better understanding of how to successfully treat their patients. The number of human lives that can be saved from this disease by predicting the number of total cases make this an important problem to analyze.
Before beginning the modeling process, I had to convert both the San Juan and Iquitos datasets into time series objects. For San Juan, I chose to only look at predicting 17 weeks since the last full year that was given was 2007. For simplicity, I wanted to only predict the remaining weeks after the last full year of data. Likewise for the Iquitos dataset, I decided to predict the remaining weeks after the last full year of data in 2009. Essentially, these were two different time frames that the problem required you to predict.
I enjoyed participating in this competition because of how applicable this was in real-world situations, especially since we are still currently suffering through a world pandemic. If we had known about COVID-19 earlier and foreseen it, I can only imagine how much it would have helped with preparation on how to handle the situation both government-wise and community-wise. As always, I learned that there are so many different choices of forecasting models, and that it is difficult to find the best fit. I was also shocked at how a basic regression model could outperform the other more advanced models— this makes it important not to dismiss any type of model and to consider all options when forecasting.