APPROACH:
Predictive modelling is based on historical time series including google mobility index, health statistics, other macro-economic variables like GINI index or total export from China (without any underlying distribution assumption)
• Based on the most recent estimations: Starting from this week, it is predicted that the new daily case count will continue to rise.
• For the upcoming two weeks, it is expected new daily case count will stay around 1500.
EXAMPLE VARIABLE:
One of the variables in the model is Google Mobility’s ‘Work Places Change from Base Line”. This shows how the number of visitors in the workplaces has changed compared to reference days*. (Or the change in how much time is spent in these places)
As we know, in the last 7 days, daily new case count started to increase as well as this variable. It might be still early to tell but since there is a significant increase for this variable in June, we expect the number of cases will increase accordingly.
*https://www.google.com/covid19/mobility/data_documentation.html?hl=tr
PREDICTIVE MODELLING PROCESS
1. For each country, day in which cumulative case count exceeds 100 is taken as pandemic start day.
2. For each country, Loess curve fitting is applied to smooth out the daily new case counts.
3. Predictive models are built to predict next ith day for i is from 1 to 12.
4. Some candidate input features are Google’s mobility features, physician counts, hospital bed counts, specialist surgical workforce, labor force, life expectancy at birth, smoking prevalence, GINI index
PERFORMANCE
Performance is measured on out-of-time test data, while comparing with all countries’ predictions and smoothed realized values with MAPE statistics (mean absolute percentage error)
Average MAPE for 1 to 12 day prediction: 10.5%
You can follow our English language YouTube videos @ REAL TURKEY: https://www.youtube.com/channel/UCKpFJB4GFiNkhmpVZQ_d9Rg
And content at Twitter: @AtillaEng
Facebook: Real Turkey Channel: https://www.facebook.com/realturkeychannel/