資料探勘技術於不孕症問題之分析與應用

性開放、人工流產合法化、晚婚、工作壓力、..等因素，以致我國每7~10對夫妻當中，就有一對不孕症症狀。不孕症是人類生殖系統的一種疾病；若夫妻在一年內有規律的性行為，且無避孕的情形下仍無法受孕時就可稱為不孕。本研究運用行政院衛生署國民健康局之人工生殖資料庫探勘國人的人工生殖資料，建立一套人工生殖資料預測模型。並且探討人工協助生殖資料庫中不孕症之重要因子，並期望在未進行手術前，由本系統提供醫師術前評估及人工生殖手術方法之選用，以提昇懷孕成功率。以往婦產科醫師針對不孕病患進行診斷時，大部份皆依據各醫生之經驗做出決策判斷。本文以醫師經驗法則、判別分析(DA)、羅吉斯迴歸分析(LOGISTIC)、…等方法，對醫師經驗法則資料進行重要屬性挑選，透過粗略集合理論(RST)、判別分析(DA)、…等，比較分類之結果。經實驗分析後，針對懷孕結果預測、生產結果預測、新生兒體重預測，其準確率皆有九成以上之準確度。以期能提供給醫師不孕症患者相關診斷時的的決策參考。

關鍵字

資料探勘；判別分析；羅吉斯迴歸分析；逐步前饋式搜尋法；因素分析；粗略集合理論；倒傳遞類神經網路；人工協助生殖

並列摘要

In Taiwan, factors such as sex liberation, legalization of artificial abortion, late marriage and work stress have caused infertility in one out of every seven to ten couples. Infertility is a disease of the human reproductive system; it is defined as the incapability in achieving pregnancy in spite of determined attempts by heterosexual intercourse without contraception within a one year period. This research employs the artificial reproduction database from the Bureau of Health Promotion (Department of Health, R.O.C.) and explores the artificial reproduction data of the general public. In addition, a set of data prediction model for artificial reproduction is established and the significant factors of infertility in the assisted reproduction database are investigated. This system is expected to provide the physicians with preoperative analysis and decision for surgical procedures in assisted reproduction, which would improve the conception rate. Previously, obstetricians and gynecologists generally make decisions based on their previous experience when diagnosing infertile patients. This research adopts methods including doctor’s rule of thumb, Discriminant Analysis and Logistic Regression Analysis to select the important attributes from the data for doctor’s rule of thumb. Subsequently, the results for the categorization are compared through Rough Set Theory (RST), Discriminant Analysis (DA) and etc. After the experimental analysis, the accuracy rate for pregnancy prediction, reproduction prediction and infant weight prediction are above ninety percent. Moreover, this research is expected to provide reference for physicians when they are making decision on related diagnosis for infertile patients.

並列關鍵字

Data Mining ； Discriminant Analysis(DA) ； Logistic Regression Analysis ； Sequential Forward Selection(SFS) ； Factor Analysis(FA) ； Rough Set Theory (RST) ； Back-propagation Neural Network ； Artificial Assisted Reproduction