Machine Learning-based Decision Prediction in Turkish Legal Texts

Öz

Artificial intelligence (AI) applications are becoming increasingly popular in the field of law and expanding its range of applications. In this study, machine learning supervised learning models are used to predict the decision type (Partial Recommendation, Partial Rejection, Recommendation, Rejection and Partial Recommendation) of the Ombudsman's Office (Ombudsman) decisions. Supervised learning models such as Decision Tree (DT), Support Vector Machine (SVM), Naïve Bayes, K-nearest neighbors (KKN), Logistic Regression (LR) and XGBoost were used in the study. During the training of the models, the “APPLICANT'S CLAIMS AND DEMANDS” section of the decision text was taken into account and the other parts were not shown to the models. Texts were transformed into TF-IDF vectors and the models were trained. The results obtained and the performance of different models are compared and analyzed in detail. It was observed that SVM was the most successful model in predicting the Ombudsman decisions on the test dataset. The SVM model achieved 61% F1 score and 62% accuracy in decision prediction.