Ongoing project / Led by Prof. Cho-Ying Huang
Within this theme, we explore challenges to nature and humanity across East Asia, including: (1) human adaptations to environmental change and their impact on human development; (2) the impact of human impacts on the environment with an emphasis on biogeochemical cycles; and (3) studies on political ecology, such as environmental governance, indigenous community governance, the tea industry, and border management, to bridge the juxtaposition of nature and humanity.
Achievement
Developing Machine Learning Algorithms for Tracking Epidemics in Time and Space
Epidemic diffusion is a space–time process, which can be considered as the movement of linked cases through space and time. Therefore, space-time locations of cases are key to identify any diffusion process. Dr. Tzai-Hung Wen, Professor at Department of Geography, National Taiwan University, borrowed the concepts of tracking the routes of typhoons for developing machine learning algorithms for profiling diffusion dynamics of disease clustering and epidemic propagation. The novel algorithms that utilize the temporal lag within the diffusion process and the spatial distance between cases to detect the spatial-temporal sub-clusters and to uncover the development of progression chains. The progression of dengue epidemics in south Taiwan from 1998 to 2015 are used for demonstrating the capability of the algorithms. Dr. Wen's work contributes a more detailed and in-depth understanding of the geographic diffusion process of epidemics. These results have been published in the international journals, including Scientific Reports 7:12565 and Annals of the American Association of Geographers 108(4):1168-1186.
Dr. Wen suggests that health authorities can integrate the epidemic reporting system with the algorithms as an automatic early-warning decision tool for uncovering the evolution of disease transmission in time and space, such as dengue, measles and tuberculosis. Currently, the algorithms are adopted in National Mosquito-Borne Diseases Control Research Center at National Health Research Institutes (NHRI) and the local government’s Department of Health for tracking the epidemic progression of dengue fever outbreaks this September.
We propose a new classification of disease cluster evolution (Figure 1) separated into single and multiple disease clusters. The changes in the center and the area of a transmission cluster were used to monitor the single pattern. Each of these changes implies a significant property of diffusion, namely, that distinct strategies for public health are needed. For example, if the epidemic in a transmission cluster becomes more prevalent than before, the area might get larger. The health authority could focus its attention there. If a transmission cluster moves to another place, its center could move. The health authority could track its trajectory and try to stop its progression. By combining the changes in these two characteristics, six distinct types of single cluster evolution patterns could be discovered. Furthermore, characterizing interaction cluster evolution patterns focuses on the changes resulting from the interaction between two or more transmission clusters.
To further prove the effectiveness of our algorithm, Figure 2 shows that our approach is more intuitive than the ST-KDE. Our approach, MST-DBSCAN algorithm, can further demonstrate a dominated diffusion type in each hot spot of ST-KDE. It includes that Hot spot I is the earliest one among these four hot spots, and its location is zone 1. Hot spots II and III are located in zone 4, and hot spot IV is located in zone 5. The emergence of hot spot IV occurs later than that of the previous two hot spots.
Figure 1: Types of cluster evolution
Figure 2: Comparing the results of MST-DBSCAN algorithm with space–time kernel density estimation (ST-KDE).