Comparison of Daily COVID19 cases spatial prediction models based on features' importance and analysis of related mobility restriction measures taken by the Government of India

The Team

Poojan Vachharajani, Mohit Chaurasiya, Arnav Gupta, Vinamra Harkar

Supervision & Support: Prof. Luo Wei, National University of Singapore

Introduction

This project implements various models to predict daily COVID-19 cases primarily using spatial data provided by Facebook's (Meta's) Data For Good program. Various Machine learning and Deep learning models are compared on the basis of the accuracy of predicted values. This project also justifies the importance of several features such as daily COVID testing statistics and daily vaccination statistics. We intent to evaluate union and state governments' policy decisions around mobility restrictions and lockdowns in Maharashtra, based on our COVID-19 spread prediction results. We found a strong significance of features like daily testing and vaccination statistics for better performance.

Data

Population Movement and Population Density data was taken from Facebook (Meta) Data For Good Portal (https://dataforgood.facebook.com/) and the COVID-19 data was taken from (https://www.covid19india.org/).

Results for Maharashtra state

Model	Feature/s	RMSE
Linear Regression	Spatial	6350.93
Linear Regression	Spatial+Vaccinations	7426.58
Linear Regression	Spatial+Tests	8528.57
Linear Regression	Spatial+Vaccinations+Tests	7735.01
Logistic Regression	Spatial	1383.73
Logistic Regression	Spatial+Vaccinations	1324.25
Logistic Regression	Spatial+Tests	1383.73
Logistic Regression	Spatial+Vaccinations+Tests	3370.27
Random Forest Regression	Spatial	1765.45
Random Forest Regression	Spatial+Vaccinations	2227.77
Random Forest Regression	Spatial+Tests	1583.89
Random Forest Regression	Spatial+Vaccinations+Tests	2414.82
RNN with LSTM	Spatial	1242.15
RNN with LSTM	Spatial+Vaccinations	1343.72
RNN with LSTM	Spatial+Tests	899.97
RNN with LSTM	Spatial+Vaccinations+Tests	1039.67

Summary

• RNN LSTM models, best with Tests as an additi onal feature.

• Logistic Regression models, best with Vaccination as an additional feature.

• Random Forest Regression models, best with Tests as an additional feature.

• Linear Regression models (benchmarking).