부산대학교 도서관

The spread of COVID-19 prediction is one of the very important tasks in determining quarantine policies in terms of a pandemic. With the recent development of artificial intelligence, the use of various prediction methods has been proposed. In addition, various data on the spread of COVID-19 are provided by each country. However, despite abundant data availability because of different prevention policies and situations in each country, the use of data available through machine learning to predict COVID-19 situations in certain countries is still limited. To leverage sufficient data for the prediction using deep learning, we attempted to make predictions via regional spread data in a country for weekly predictions of COVID-19 growth over time using long short-term memory (LSTM) methods. In addition, the numbers of confirmed cases, recoveries, and deaths, and the variable population were noted to track the correlation with the accuracy of prediction. The accuracy assessment of the presented model was based on root mean square error, mean absolute percentage error, and graph visualization. We believe that achieving it accurate predictions using biased data, such as a sharp increase in the numbers of confirmed cases because of highly contagious variants, including omicron variants, is not easy. However, LSTM can predict patterns similar to that of actual data using various variables and regional data.