登革熱是爆發在熱帶和亞熱帶等區域的傳染性疾病，在這些區域不論是城市和郊區都有極高的染病機率，過去各爆發疫情的國家對登革熱防疫皆投入巨大的成本。本研究使用典型的空間自我相關分析驗證登革熱疫情位置具備空間相關性，並發展了空間群集分析和空間時序擴散分析兩個地理探勘技術，空間群集分析結合K-Means演算法和空間自我相關分析的結果利用可掌握的考量因素中找出哪些地理位置疫情特性相同，並以此依據進行擴散預測；空間時序擴散分析使用疫情病例資料產生空間時序資料，接著利用關聯方法的序列樣式分析找出疫情在空間的擴散路徑，利用此結果進行擴散預測。利用空間群集分析和空間時序擴散分析兩個方法在臺灣高雄地區登革熱疫情擴散區域預測的地理位置命中率為59.29%和68.40%，地理位置覆蓋率為91.46%和38.75%，皆高於一般的防疫進行方式；若作業成本和錯誤成本一樣的情況下，基於空間群集分析和空間時序擴散分析的防疫策略在成本上分別比一般防疫策略節省27.21%和47.28%，本研究提供了具備地理位置相關的疾病一個績效良好的疫情擴散預測模型，將此應用在實際防疫策略亦具備較好的成本控制效果。;Dengue fever is a contagious disease that breaks out in tropical and subtropical regions. In these regions, both cities and suburbs have a high probability of infection. In the past, countries with outbreaks have invested huge costs in dengue epidemic prevention. This study uses classical spatial autocorrelation analysis to verify that the location of the dengue fever epidemic is spatially correlated, and develops two spatial data mining techniques, spatial clustering analysis and spatial time sequence diffusion analysis. Spatial clustering analysis combines the results of K-Means algorithm and spatial autocorrelation analysis to find out which geographic locations have the same epidemic characteristics from available consideration factors and use this basis to predict the spread region. Spatial time sequence diffusion analysis uses epidemic case data to generate space-time sequence data, then use the sequence pattern analysis to find the spread path of the epidemic in space and use this result to predict the spread region. Using the two methods of spatial clustering analysis and spatial time sequence diffusion analysis, the hit rates of the dengue fever epidemic area in Kaohsiung, Taiwan are 59.29% and 68.40%, and the coverage rates are 91.46% and 38.75%, which are higher than general epidemic prevention. If the operating cost and the error cost are the same, the epidemic prevention strategy based on spatial cluster analysis and spatial time sequence diffusion analysis saves 27.21% and 47.28% in cost, respectively, compared with general epidemic prevention strategies. This research provides a well-performing epidemic spread prediction model with geographically related diseases. This application research also has a better cost control effect in actual epidemic prevention strategies.
