A New Explainable AI Approach to Legal Judgement Prediction: Detecting Model Uncertainty and Analyzing the Alignment Between Judges and Models

In this study, a binary classification problem is studied for the predicting outcomes of Turkish Constitutional Court decisions with deep learning models. New approaches have been developed to calculate the uncertainty in the outputs of the models and to analyse the level of alignment of the models with the behavioural patterns of the judges. For this purpose, the performance level of the model was calculated for the decisions of the Turkish Constitutional Court with and without dissenting votes, and the uncertainty of the model outputs was analysed for these decisions with the developed approaches.