Smart justice framework using artificial intelligence for ECHR article violation classification and case outcome prediction

Abstract

This study aims to build and analyse an Artificial Intelligence-driven pipeline for automating the analysis of legal cases in the European Court of Human Rights (ECHR), to address transparent, ethical, and responsible decision support systems in the legal domain. The study applies a Natural Language Processing (NLP) framework to a dataset of 4,178 legal cases in the ECHR related to Articles 3, 5, 6, and 8. The pipeline integrates multi-label classification of article violations in legal cases using Multinomial Naive Bayes, Random Forest, and BERT. SpaCy models were used for Part-of-Speech tagging and Named Entity Recognition (NER). Latent Dirichlet Allocation (LDA) and BERTopic for topic modelling of the legal texts. Summarisation of these large legal cases was applied using a fine-tuned BART-large-CNN, and generation of case outcome prediction was done using LSTM and bidirectional LSTM that contain attention mechanisms. In multi-label classification, Random Forest achieved the highest accuracy of 64%, where Multinomial Naive Bayes scored 59%, and BERT scored 52%. While BERTopic found eight semantically coherent thematic clusters within the dataset. For case summarisation, BART-large-CNN achieved a 45% ROUGE-1 score and a 24% ROUGE-L score. In case of outcome prediction (text generation), the Bidirectional LSTM with attention scored a ROUGE-L score of 17% compared to the LSTM with an 8% ROUGE-L score. Findings from the framework demonstrate the potential of NLP-based approaches in automating legal case analysis, offering a research foundation for developing AI-assisted systems that support article violation detection, case summarisation, and evidence-based jurisprudence in the legal domain.