Predicting Criminal Cases Of Oromia Supreme Court Using Machine Learning

Abstract

In This World, Every Victim Needs A Fair And Unbiased Judicial Decision. A Judicial Decision Is A Process Of Deciding According To The Law On The Guilty Person For The Committed Crime.This Process Is Carried Out By Judges. Judges Are Human Beings; They Can Be Inclined Into Some Groups Or Individuals Due To Various Reasons. In Addition To This Problems, With Human Judicial Decision-Making, There Is A Low Similarity Of Their Judgment Among Different Judges On The Same Cases. Applying A Machine Learning Prediction Model Would Improve Decision-Making Quality And Efficiency By Automating The Process Based On A Real Dataset. This Study Aimed To Predict A Judicial Decision Of The Oromia Supreme Court(OSC) Using Machine Learning Techniques. The Judicial Decision Outcomes Would Be Predicted From Two Aspects: The Judgment (Identifying Whether The Suspect Committed The Alleged Crime Or Not) And The Penalty (If The Suspect Is Found Guilty Of The Alleged Crime, Impose Penalty) Using Criminal Case Dataset Collected From OSC. The New Dataset Has 1638 Instances That Were Used To Train The Models. This Dataset Hasn't A Balanced Instance Class.The Synthetic Minority Oversampling Technique Was Applied On Training Dataset And Generated 1736 Data To Balance. An Experimental Approach Was Performed In This Study To Determine Best Model. Machine Learning Models Based On Support Vector Machine (SVM), Random Forest (RF), And Na??Ve Bayes (NB), With Feature Extraction Techniques Like Term Frequency Inverse Document Frequency (TF-IDF), Bag Of The Word (BOW), And N-Gram Have Been Experimented For Both Judgment And Penalty Prediction Separately Using 10 Fold Stratified Cross-Validation. Different Classification Metrics Are Used For Evaluating Models.The RF Model With TF-IDF Performing Best Than Other Models For Predict Judgment. Sinceit Scored An Accuracy Of 98.5 % And An F1-Score Of 98 %. The SVM Model With TF-IDF Performing Best Than Other Models For Predict Penalty With An Accuracy Of 79.68 %, And An F1-Score Of 79 %. Finally, The Best-Performed Model Was Evaluated By Legal Experts And Achieved 77.5% Accuracy. Therefore, RF And SVM Models With TF-IDF Feature Extraction Are Recommended To Effectively Predict The Judicial Decisions Of OSC. This Study Contributes To Developing And Evaluating Machine Learning Models To Predict Judicial Decisions With Penalties. However, It Would Be Better If In Depth Analysis With All Types Of Law On Big Corpus Will Be Experimented With Deep Learning Approach For Better Results.