隨著經濟環境的不斷演進,社會架構的變遷腳步日益紊急,大眾生活條件與背景的差異亦不可同日而語,自此而衍生的犯罪型態與隱患,未可悉數;加諸行政資源有限、司法審議與裁判制度調整之研擬費時,相關執法人員將無法及時因應當前犯罪率逐年攀升之治安困境。 近年刑事政策採取寬嚴並進之精神,對於罪犯之處罰,尤以重視犯罪之環境與動機,對於偶蹈法網者,除危害國家社會法益情節重大及怙惡不悛者外,提供悔改向上之機會。在完整配套的執法及具限度之條件下,對於輕罪者施予減刑,並輔以觀護監督及更生輔導等措施,力求將減刑政策施行下所可能招致之負面反應或風險性,降至最低;然於長期密切追蹤之調查記錄下,減刑再犯議題仍不容輕忽。本研究著眼於減刑審查資料之主觀語意轉換,將知識發現之過程,得以運用資料探勘與網路學習機制,以屬量指標的最佳化呈現之。透過倒傳遞類神經網路嘗試架構一預測系統,以全國刑案大宗類別為試驗施行對象,進行犯罪資料萃取,藉以針對當前減刑犯之再犯概率做出準確分析,期使為我國執法人員帶予相關決策規劃與制定之參考。
With the evolution of the economic environment, the ever-changing social environment makes our living conditions and the differences of people’s class origin cannot be mentioned in the same breath. Therefore, numerous crime patterns and potential problems are deriving from them. With the limited administrative resources, time-consuming adjustment in the judicial review and referee system, law-executors will be unable to respond the security dilemma – a steadily rising crime rate – timely. In recent years, Taiwan’s criminal policies emphasize to strike the balance between justice and mercy. In terms of the punishment of criminals, situation and motivation have been given substantial attention. As regards those who go astray – except for someone jeopardizing public legitimate rights and interests with a significant society cost, the authorities will provide another chance to reform themselves. Under the premise that law enforcement possesses coordinated sets of measures adequately, commutation is suitable for use in misdemeanors. With probation supervision and rehabilitation counseling, law-executors can strive to minimize the negative reactions and the possible risks incurred after the implementation of commutation policy. However, based on a series of recent extended surveys, recidivism issues concerning commuted prisoners may not careless and indiscreet. This study focuses on subjective semantics conversion about the document inspection of commutation. With the application of data mining and network learning, optimized quantitative variables will present the whole process of knowledge discovery., We made an attempt to design a prediction system by using Back Propagation Neural Network (BPN) first; then, giving the examination of nationwide criminal cases – especially several misdemeanors – to implement crimical data extraction. Through this way, we hope to make an accurate analysis of recidivism rate for commuted prisoners; and to be more precisely, we aim to raise some valid arguments for the authorities to make related decisions properly.