| dc.description.abstract | Previously many studies have aimed at predicting the trend of a disease through time series forecasting using machine learning methods. However, data extracted from the real world is often noisy, which can pose numerous challenges for directly predicting the trend, and therefore leading to suboptimal prediction results. Furthermore, real-world data is usually very large, that is, having very long time periods. When it comes to data of such scale, trend forecasting becomes intractable even to state-of-the-art forecasting algorithms such as RNN-LSTM. In the past, not much research has been conducted in applying anomaly detection for disease outbreak detection, including the most recent COVID-19 pandemic. Consequently, in this research, we propose redefining the problem into outbreak detection, which aims to predict whether a future point is or is not a sign of a large scaled COVID-19 outbreak. Through simplifying a complex regression problem into a binary classification problem, the requirements of the learning model may be decreased and therefore the learning performance may be enhanced. | en |
