● 中文摘要：

　　近年家庭暴力日趨嚴重，而家庭暴力再犯的事件也越來越多，不僅造成被害人家庭生命財產無可彌補的損失，更嚴重威脅著社會大眾的安全，因此防範家庭暴力再犯的發生，是一件非常嚴肅及重要的課題。本研究使用羅吉斯迴歸、決策樹與類神經網路方法來建立預測再犯模型，其目的在於預測再犯情形，並探討再犯者的特性。本文蒐集自2004年1月起至2009年12月止，臺灣地區各地檢署執行裁判確定之家庭暴力犯罪人資料，共計7,617筆。研究結果發現類神經網路模型，於不同的分割點下，均有較佳的預測再犯正確率，決策樹模型之再犯判別能力則較羅吉斯迴歸模型與類神經網路模型稍差。至於模型效力驗證的ROC曲線方面，羅吉斯迴歸、決策樹與類神經網路模型的ROC比率，亦顯示類神經網路模型預測效力最好。

● English Abstract:

     In recent years, events of domestic violence have increased and have serious effects on victims. The domestic violence is not only do damage irreparably to the lives and property of victims, but it also threaten safety of our society. As a consequence, how to prevent the recidivism of domestic violence becomes a very important issue. This study aims at developing forecasting models to predict the possible occurrence of recidivism. These models include logistic regression, decision tree and neural network models. In this study, a sample of 7617 people who were confirmed to guilty domestic violence crimes from January 2004 to December 2009 was used for demonstration of the effectiveness of the forecasting models. In addition, the characteristics of the recidivist are investigated. The research findings indicate that the neural network model has the best forecasting capability of recidivism at different divided points when compared with other two models. The decision tree has the worst forecasting capability among all of the models. As for measuring the effects of models, the ROC rate also shows that the neural network model is the best forecasting method.

● 文章連結：

https://hdl.handle.net/11296/yj8364

● 資料來源：

臺灣博碩士論文知識加值系統