應用類神經網路系統建構妨害性自主罪受刑人再犯預測模型

Using Artificial Neural Network Model for the Sexual Offender Recidivism Prediction

劉昌誠(Chang-Cheng Liu)

指導教授 ： 徐建業

臺北醫學大學/醫學院/醫學資訊研究所/碩士(2010年)

https://doi.org/10.6831/TMU.2010.00178

摘要

我國的法律規定妨害性自主罪受刑人需接受專業鑒定，但卻缺乏「高再犯可能」的標準。本研究期待利用類神經網路預測模型，提供精神醫療人員進行司法鑒定時重要的決策支援，進而加強預測度、降低與實際的誤差，同時節省人力、財力。本研究以民國84及85年間自臺灣北部某監獄出監的552位個案，分別追蹤其自離開監獄至92年12月31日及93年12月31日止的妨害性自主罪再犯情形為樣本。分析相關性較高的22個再犯因子後，利用類神經網路建構預測模型，並與邏輯迴歸模型、RRASOR、MnSOST-R及Static-99的ROC曲線相比較以檢驗其預測能力。建構出的類神經網路模型擁有最大的ROC曲線之曲線下面積，和其他預測方法相比整體而言有較佳的預測能力。

關鍵字

妨害性自主 ； 再犯預測 ； 類神經網路

並列摘要

In Taiwan, sexual offenders need to receive professional assessments before leaving the prison, but the lack of "high recidivism risk" standard is a serious problem. This study try to use artificial neural network model to provide an important reference to forensic psychiatric professionals, thereby strengthening the predict ability, reduce errors, and saving manpower and money. Participants of this study were 552 sexual offenders released from a prison in northern Taiwan in 1995 and 1996, and we follow all cases from the time of release to December 31, 2003 and 2004 separately. 22 risk predictors with statistic significance are selected to construct a artificial neural network (ANN) model for the sexual offender recidivism prediction. Then we examined the predict ability of the ANN model by receiver operating characteristic (ROC) analysis, and compare with the model constructed by logistic regression, RRASOR, MnSOST-R, and Static-99. The area under the ROC curve for ANN model is the biggest, after comparing with all prediction models, ANN model got better predict ability.

並列關鍵字

sexual offense ； prediction of recidivism ； artificial neural network

參考文獻

[10]林明傑、董子毅、蔡昇運、陳珍亮（民94）。台灣性罪犯靜態再犯危險評估量表(TSOSRAS)之建立及其外在效度之研究。亞洲家庭暴力與性侵害期刊，1(1)，49-110。

[17]電腦處理個人資料保護法。

[13]黃健、溫瑞祥、黃永順、鄭安雄（民94）。性犯罪再犯風險預測因子的本質與應用。亞洲家庭暴力與性侵害期刊，1(1)，137-162。

[20]Epperson, D., Kaul, J., Huot, S., Goldman, R., & Alexander, W. (2003). Minnesota sex offender screening tool–revised (MnSOST-R) technical paper: Development, Validation, and Recommended Risk Level Cut Scores. Ames, Iowa: Iowa State University. [Online] available at http://www.psychology.iastate.edu/~dle/MnSOST-RManual12-22-2005.pdf

[21]Florio, T., Einfield, S. and Levy, F. (1994). Neural networks and psychiatry: candidate applications in clinical decision making. Australian and New Zealand Journal of Psychiatry, 28(4), 651-666.