Using Dynamic Adjusting NGHS-ANN for Predicting the Recidivism Rate of Commuted Prisoners

Po-Chou Shih,Chi-Hsun Chou,Chui-Yu Chiu

doi:10.3390/math7121187

Abstract

Commutation is a judicial policy that is implemented in most countries. The recidivism rate of commuted prisoners directly affects people’s perceptions and trust of commutation. Hence, if the recidivism rate of a commuted prisoner could be accurately predicted before the person returns to society, the number of reoffences could be reduced; thereby, enhancing trust in the process. Therefore, it is of considerable importance that the recidivism rates of commuted prisoners are accurately predicted. The dynamic adjusting novel global harmony search (DANGHS) algorithm, as proposed in 2018, is an improved algorithm that combines dynamic parameter adjustment strategies and the novel global harmony search (NGHS). The DANGHS algorithm improves the searching ability of the NGHS algorithm by using dynamic adjustment strategies for genetic mutation probability. In this paper, we combined the DANGHS algorithm and an artificial neural network (ANN) into a DANGHS-ANN forecasting system to predict the recidivism rate of commuted prisoners. To verify the prediction performance of the DANGHS-ANN algorithm, we compared the experimental results with five other forecasting systems. The results showed that the proposed DANGHS-ANN algorithm gave more accurate predictions. In addition, the use of the threshold linear posterior decreasing strategy with the DANGHS-ANN forecasting system resulted in more accurate predictions of recidivism. Finally, the metaheuristic algorithm performs better searches with the dynamic parameter adjustment strategy than without it.

Highlights

Parole is the temporary and conditional release of a prisoner prior to the completion of their maximum sentence period
In order to verify the performance of the dynamic adjusting novel global harmony search (DANGHS)-artificial neural network (ANN) forecasting system, we compared the extensive experimental results of DANGHS-ANN with five other systems, including various harmony search (HS)-ANNs and one backpropagation network (BPN)
We referred to previous references [17,20,21,22] and used the trial and error method to decide the parameters of different HS algorithms

Summary

Introduction

Parole is the temporary and conditional release of a prisoner prior to the completion of their maximum sentence period. Commutation is the substitution of a lesser penalty for that originally given at the time of conviction. Whether on parole or commutation, if the prisoner reoffends it can cause social disruption. This highlights the need for accurate recidivism predictions for parolees and commutation offenders. Carroll et al [1] stated that “in Pennsylvania, the parolees with alcohol problems, younger parolees, and those originally convicted of property crimes (rather than assaultive or drug crimes) were more likely to commit new crimes on parole. Offenders with past heroin use were convicted of more serious crimes on parole. Absconding was significantly more predictable for cases with prior convictions, previous parole violations, and miscellaneous negative statements by the institution about the inmate’s personality.”. Absconding was significantly more predictable for cases with prior convictions, previous parole violations, and miscellaneous negative statements by the institution about the inmate’s personality.” In Williams’ paper [2], they demonstrated that in California, non-sex offenders, drug registrants, offenders with more than one felony conviction, Mathematics 2019, 7, 1187; doi:10.3390/math7121187 www.mdpi.com/journal/mathematics

Methods

Results

Conclusion