自民國84年3月1日起臺灣開始實施全民健康保險的全民醫療保險政策後,人民的壽命大幅的增加。但好的醫療體系除了事後的治療以外,更需要良好的事前預防,因此,本研究將透過粒子群最佳化演算法結合分類器有效預測疾病的產生及影響疾病的重要變數,並期望本研究成果可以為醫療預防做出貢獻。本研究將以產後憂鬱症為例,由於產後憂鬱症是許多婦女產後最常併發的病症之一,產後憂鬱症除了會影響婦女本身之外,更會影響新生兒的身心健康與發展,因此,產後憂鬱症為近年來受重視的精神疾病。本研究期望能利用全民健康保險研究資料庫的資料對產後憂鬱症進行研究。但由於全民健康保險研究資料庫有一最主要的缺點:類別不平衡問題,即大多數的樣本為未罹患疾病的病人,因此,產生罹患疾病及未罹患疾病的病人比例過於懸殊。為了解決上述問題,本研究將先利用分層抽樣降低類別不平衡的影響,再找出最適當的分類器,最後再結合粒子群最佳化演算法偵測產婦罹患產後憂鬱症,進而搜尋出重要變數。
Since Taiwan implemented national health insurance policy on March 1, 1995, people's lifespan increased significantly. However, a good medical prevention is more importantly than a good medical treatment. Accordingly, this paper wants to predict the disease and to search the relevant variables for medical prevention by using particle swarm optimization and classifier. This paper uses the proposed model to predict postpartum depression because postpartum depression is one of the most common complications of women. The postpartum depression might affect women herself and affect the development of newborns. Therefore, the postpartum depression is an important mental illness in the recent years. This paper uses National Health Insurance Research Database to predict postpartum depression. However, National Health Insurance Research Database has an important disadvantage, namely, class imbalance problem. To solve this problem, this paper uses stratified sampling to reduce the effect of class imbalance problem and then search the best classifier for postpartum depression. Finally, particle swarm optimization combines the searched best classifier to predict postpartum depression and to search the relevant variables.