Download This PaperA Network-Based Approach to Predict New Affected Regions and the Spread Evolution of COVID-19
16 Pages Posted: 21 Apr 2020
Date Written: April 16, 2020
Abstract
Given the most recent events involving the fast spreading of COVID-19, policy makers around the world are being challenged with the difficult task of developing efficient strategies to contain the dissemination of the disease among the populations, sometimes by taking severe measures that restrict the local activities both socially and economically. Within this context, models which can help on predicting the next regions to be affected by the disease and also its spread evolution in a specific region would surely help the authorities on their planning. However, the current prediction attempts in this sense are usually either for an unspecified time-range or by utilizing distinct models, focusing on a specific region. In this paper, we introduce two different network-based models by making use of preliminary available data regarding the spread of COVID-19. The first model predicts the new regions to be affected by the disease within a certain time range. It starts by mapping each region as a node in a network, and the edges are generated according to their proximity in terms of geographic coordinates. Afterwards, we apply link prediction techniques for generating the predictions within a predetermined number of days. The obtained experimental results on this task achieved an average accuracy of 90%, when predicting the next 10 regions to be affected within the next 21 days. The second model proposed in this work predicts the evolution of time series through temporal networks. In this case, each node represents a time series, and the edges are created according to the similarity of their variations at each time step. The results obtained by applying this model on time series concerning the spread evolution of COVID-19 on different world regions are promising, with the predictions being consistent with later real spread evolution data released for these same regions.
Keywords: complex networks, machine learning, classification, time series prediction, temporal networks, Covid-19
Suggested Citation: Suggested Citation