Improving Accuracy in the Montgomery County Corrections Program Using Case-Based Reasoning

Abstract

The Montgomery County Corrections Program is a program designed to address the problem of overcrowded jails by providing an out-of-jail rehabilitative program as an alternative. The candidate offenders chosen for this program are offenders convicted on nonviolent charges and are currently chosen subjectively with little statistical basis. In addition, historical data has been recorded on offenders who have passed through the program, making the program a good candidate for case-based reasoning. Using such reasoning, County Officials would like an objective measurement which will predict the success or failure of a candidate offender based on past offender history. The four case-based reasoning algorithms chosen for this prediction are Discrete, Continuous and Distance Weighted k-Nearest Neighbors and a General Regression Neural Network (GRNN). Although all four algorithms prove to be an improvement on the current system, the GRNN performs the best, with an average accuracy rate of 68%.