Fairness in Predicting Recidivism Score

The growing use of machine learning models in many domains has made human activities much simpler. One of the key domains in which machine learning models are applied is criminology and criminal justice. The prediction of recidivism is the main topic of this paper. Recidivism is defined as an individual reoffending a crime after being released on bail, sentence, or parole. Predicting the likelihood of an individual reoffending after involved in criminal justice is called recidivism score. Several algorithms exhibit bias related to age, gender, and ethnicity, among other factors that affect fairness in score prediction. To address this, bias migration and dimensionality reduction are used during the pre-processing stage of a dataset. The Correctional Offender Management Profiling for Alternative Sanctions (COMPAS) dataset ( https://www.kaggle.com/datasets/danofer/compass/data [1]; Larson et al. in How we analyzed the COMPAS recidivism algorithm. ProPublica, p 9, 2016 [2]), collected by ProPublica and obtained via Kaggle, is used. Several models are used to train the dataset and compare how this model find the hidden patterns and reduce bias in predicting, and they are support vector machine (SVM), decision tree, and ensemble learning. The accuracy of the models is SVM 67.85%, decision tree 66.55%, and ensemble learning 68.01%.