Classification of Criminal Recidivism Using Machine Learning Techniques

Abstract

There are numerous cases in the recent times, where a criminal commits a crime, immediately after being granted parole, this is called Criminal Recidivism. The act of recidivism poses a great threat to the society and thus needs to be checked. This paper posits a machine learning approach to detect and predict the tendency of a criminal to commit recidivism. The proposed system helps classify the criminals into Low, Medium, and High risk of committing recidivism. Features like ‘Ethnic code’, ‘Marital Status’, ‘Age’, ‘Sex Code’, ‘Legal Status’ and many more are considered while training the model on the dataset. Supervised Classification Algorithms are implemented, and voting is subsequently done, to select the algorithm with the highest accuracy. The Random Forest Algorithm provides the highest accuracy score followed by KNN and lastly Logistic Regression. Moreover, the data is analyzed using visualization charts, where various attributes are deeply analyzed in relation to the target variable ‘Score Text’. Graphs between these attributes and the target variable highlight trends, which may provide useful insights to parole granting authorities while assessing a criminal for parole. Stratified K-Fold Cross Validation is used to bolster the results of the algorithms, which gives us accuracy score similar to the above algorithms. Thus, it validates and renders the algorithms unbiased and fair.