因為登革熱需要二次感染才會造成重症的特性，死亡率不高，造成疫情的嚴峻程度較不起眼，但隨著全球暖化，登革熱的分布範圍漸漸改變；過去國內外對於登革熱的研究中，最常見的是以氣候因子做登革熱的預測，再來是氣候因子結合社會因子與氣候因子結合地理因子，目前並無將三種因子結合的研究，故本研究結合三種因子，與臺灣登革熱的病例資料，以人口數量劃分的二級區域做為顆粒度，利用 Random Forest 與XGBoost 建立登革熱隔週感染區域預測模型。最後實驗結果 Random Forest 與 XGBoost的 ROC/AUC 皆高於 93%，且 Recall 皆高於 80%，依照此結果，政府單位可以更精準的去判別需要噴藥撲滅的登革熱可能感染區域，進而降低人力成本與醫療資源。;Death rate of dengue fever is low, because dengue fever become severe illness only when second infection happened. However, global warming is getting server recently, which make the infection distribution of dengue fever different. Common method of previous studies use climate factors combined with social or geographic factors to predict dengue fever. However, recent study did not use combination of these three factors into dengue fever prediction. We proposed a method that combines these three factors with data of Taiwanese dengue fever and uses the secondary area divided by the population as the granularity. Random Forest and XGBoost are used for prediction model of weekly dengue fever infection area.
Experimental results showed that the ROC/AUC of Random Forest and XGBoost are both higher than 93% The Recall rate is higher than 80%. With the result, government can determine which area should do disinfection process to reduce the infection rate of dengue infection. Because of accurate prediction and disinfection process, the personnel cost can be reduced and it can prevent waste of medical recourse.
