以巨量資料分析觀點探討受保護管束者再犯風險評估工具之研究
- 發布日期:
- 最後更新日期:111-07-27
- 資料點閱次數:469
● 主管機關:法務部
● 執行機構:國立台灣科技大學建築系
● 研究期間:10807 ~ 10905
● 中文關鍵字:
受保護管束者;再犯預測;資料探勘;類神經網路;風險評估
● 英文關鍵字:
● 中文摘要:
● 英文摘要:
Statistics from the Ministry of Justice of the Republic of China show that the implementation of community-based interventions in criminal cases is the most clearly evidenced by increases in the number of parole cases. In 2006, for example, the overall number of annual parole cases being handled by the district prosecutors office was 23,807. By 2016, there were 37,584 annual parole cases being handled by the district prosecutors office -- an increase of nearly 1.6 times in 10 years. However, with respect to repeat offenders on parole who had been given a determinate sentence, the 2009 parole absolute revocation (recidivism verdict determined) rate was 7.97 percent, and by 2018 the parole absolute revocation (recidivism verdict determined) rate was 13.1 percent; compared with the number of parole revocations due to recidivism from 2009 to 2018, which increased at a faster rate than the parole release rate. Meaning the speed of parolees returning to prison was quite high. Obviously the successful reintegration of parolees into society has not been not going as expected, and the number of prisoners has been hovering at more than 55,000 for the past few years. About 70 percent of the inmates are second offenders or recidivists, and the increase in the number of people having their parole revoked and returning to prison has made the prison overcrowding and recidivism increasingly serious issues. Besides affecting societal safety and causing overcrowding in correctional institutions, parolee recidivism has had a considerable impact on probation officer workload. From 2011 to 2017 the number of officers at district prosecutor offices has ranged from 214 to 228. However, the number of individual core cases of parolee monitoring work has increased from 2,900 at the end of 2012 to 3,762 at the end of 2018, with each probation officer being required to monitor an average of 248 individuals. It can be seen from Ministry of Justice figures that probation officer’s casework load is becoming increasingly arduous as time goes on. Therefore, reducing the possibility of recidivism amongst parolees through prediction and trying to alleviate prison overcrowding is the primary orientation in research on the subject of crime prevention involving re-offending parolees. The examples of Taiwan's current application of big data in policing are from crime detection and investigation, yet there is scant research on the prediction of parolee recidivism risk or a body of research created through the construction of related evaluation tools. The main purpose of this study is to construct a set of localized recidivism risk assessment tools, research and collect relevant crime prevention documents from Taiwan and abroad to consolidate recidivism factors into one body, and utilize expert questionnaires in the application of the Fuzzy Delphi Method of statistical analysis to screen out the core recidivism factors affecting the parolees. Founded on Big Data analysis technology, the information on parolees from the Taiwanese legal system over the past decade will be used to mine and analyze the correlations between variables and recidivism factors through the associative data exploration. Thereafter, Artificial Neural Networks will be utilized to establish a tool to predict high-risk behaviours and assess recidivism risk. The research results can then assist the technical work of probation officers and community interventionists by providing an objective reference, and establish a decision-making analysis support model with which to achieve effective crime risk management and effectively reduce recidivism, thereby maintaining public safety. The study founded that: 1. In the early stage of this study, consensus among experts was reached by Fuzzy Delphi Method, and the index of evaluation for recidivism risks consisted of 6 dimensions and 13 evaluation factors. The 13 evaluation factors included: type of crime, number of times of imprisonment, criminal records, age of onset of offending, age at release, violation records in prison, record of protective control or parole revoked, age, gender, imprisonment record of family members, area of residence, record of alcoholism or abuse of drugs, as well as emotional and mental state. 2. Big data analysis showed that, the factors which showed highest correlations with recidivism of offenders under protective control included age at first expiry of protective control, age at first commencement of protective control, age, gender, number of times of imprisonment, number of criminal records, area of household registration and type of crime. The 8 factors of recidivism mentioned are worthy of special attention. 3. Cases of offenders under protective control in the last ten years were put through artificial neural system training, with high correlation recidivism factors as the input layer, and the factor of whether the offender recidivated under protective control as the output layer. 70% of the cases were used as training cases, while 30% of the cases were used as testing cases. It was verified that both the results of training and testing showed accuracies of 73%. After discussions on expert symposiums, this study has the following suggestions: 1. The data for this study was drawn from the “case management system of prosecutor office”, “prison administration system” and “criminal case system”, collected and collated by Department of Information Management, Ministry of Justice. Throughout the process of data analysis, many fields in the database were found to be in null value, in a frequency that was considered too high, and there were many typographical errors. In addition, some of the fields only contain handwritten scripts by the probation officers, and there was no tendency to digitalize the records. This study suggests to set up a relevant database for probation system, for the convenience of probation officers to input and retrieve data of offenders under protective control. Different systems of Ministry of Justice should follow a standard format to facilitate application of big data. 2. Offenders under protective control with low level of recidivism should be treated with diverse supervision and counselling measures, including education to enhance law awareness, reducing frequency of report, and allowing them to report in writing. For offenders under protective control with middle level of recidivism, on top of enhancing law awareness, they should be bound by increased frequency of report, irregular urine tests and multiple supervision. Offenders under protective control with high level of recidivism should be requested to increase frequency of report and urine test, and report to local police. The authorities should assist them to achieve job stability, securing core of lives, while implementing multiple supervision or even monitor their acts. To increase efficiency and reduce the workload of probation officers, the authorities should cooperate with the current system of volunteer probation officers and social welfare units, while enlisting help from other social resources. 3. The recidivism risk assessment tool in this study generated predictions for recidivism risks from big data, but in practice, probation officers should employ their professions and experiences to evaluate various conditions in the cases, to decide on options for better outcomes.
● 文章連結:
https://www.grb.gov.tw/search/planDetail?id=13148836
● 資料來源:
GRB政府研究資訊系統