COVID-19 has already had a devastating impact on economic and social development and people's life all over the world. Up to September 22nd, 2020, more than 30 million people have been infected. Finding out how to predict and estimate the pandemic trend precisely is of huge necessity because COVID-19 has made the world economy in a recession and deprived over 700 thousand lives. This paper is dedicated to making a comparison between conventional machine learning regression models, including ridge regression and lasso regression, and multivariate polynomial regression. Besides, we attempt to use statewide data to fit nationwide data in the US, proposing a novel aspect to forecast pandemic trend.Under the current situation, insufficient data limit the use of machine learning. To address the issue of data deficiency, a classification model based on neural network with Twitter data is applied. This gives out an alternative approach to estimate daily increase of infected people.We discovered that in a different period, specific models would outweigh another models' performance. Also, our result showed that the data of Georgia and Massachusetts could represent the whole nation data with linear transformations. And this paper verified that using alternative data that relate to the COVID-19 situation to alleviate data deficiency is feasible.
CITATION STYLE
Yang, Z., & Chen, K. (2020). Machine Learning Methods on COVID-19 Situation Prediction. In Proceedings - 2020 International Conference on Artificial Intelligence and Computer Engineering, ICAICE 2020 (pp. 78–83). Institute of Electrical and Electronics Engineers Inc. https://doi.org/10.1109/ICAICE51518.2020.00021
Mendeley helps you to discover research relevant for your work.