Using artificial intelligence to predict decisions of the Turkish constitutional court

Abstract

This article draws on an artificial intelligence (AI) technique to predict whether an individual application regarding the Turkish Constitutional Court's public morality and freedom of expression cases leding to a "violation" or a "nonviolation" decision. To this end, four different data sets have been composed, preclassification and fundamental word embeddings steps have been made on each data set. Multilayer perceptron, which is based on artificial neural networks, has been used for the prediction of the case decisions. We have predicted the court's decisions on these cases with the high success rates (average accuracy of 90%) by using the subject or reasoning sections of texts of the cases as data. The subject section of the cases constituting only a very small part of the data has yielded the highest accuracy. The article has demonstrated that a basic AI technique can be successful in achieving accurate predictions even with relatively small data sets derived from well-structured court rulings.