Convolutional neural network-based automatic prediction of judgments of the european court of human rights

In the past few years, predictive modelling has brought revolutionary changes in the way various industries function. Advancements in the areas of Deep Learning (DL) and Natural Language Processing (NLP) have made their application to different problem areas highly promising. In the legal domain, positive results have been obtained in predicting the judgements of various Courts of different countries using DL and NLP. However, not much research has been carried out in the area of legal judgement forecasting for the European Court of Human Rights (ECHR). The models designed in the previous research employ only one Machine Learning algorithm namely a Support Vector Machine (SVM) to solve such problem. This study applies DL and NLP to the problem of automatic prediction of judgements for ECHR. Extensive experiments are conducted which compare the performance of models trained on SVM with linear kernel as part of previous research (Medvedeva, Vols & Wieling, 2018) with the models trained on Convolutional Neural Networks (CNN) as proposed in this study. To implement this, state-of-the-art NLP techniques are applied to the text data. Moreover, pre-trained and custom trained Word Embedding text representations are considered. Statistical tests are performed to gather sufficient statistical evidence to determine which algorithm performs better at providing a solution to this problem. Based on the results obtained, it is established that overall, CNN models outperform SVM models as the former achieves an average accuracy of 82% whereas the latter achieves 75%. Specifically, CNN models for four Articles out of nine achieve statistically significant higher accuracy than SVM models.