Scikit-learn was utilized to predict the likelihood of a criminal reoffending, using the COMPAS dataset. Data cleaning procedures involved class rebalancing techniques such as SMOTE and filling in null values. Various classical machine learning models were experimented with before ultimately settling on the random forest model, with a 93.5% accuracy result.