Transformer-based embeddings and stacking classifiers for judgment outcome prediction in multilingual legal texts

Legal systems in many Commonwealth countries operate primarily in English, even though citizens speak various regional languages. This language barrier often hinders access to justice, making legal information and services inaccessible to those not proficient in English. This study addresses this critical issue by developing a novel approach to predicting legal outcomes using summarized legal texts in both English and regional languages in India, namely, Kannada, Tamil, and Telugu, aiming to bridge the linguistic divide in the Indian judicial system. The study demonstrates that summarized texts and their regional language versions can achieve effective legal outcome prediction (LOP). The proposed system, which includes an embedding module with RoBERTa, FastText, InLegalBERT and IndicBERT, and a multi-stage classification module, employ classifiers such as Random Forest (RF), Decision Trees (DT), Support Vector Machines (SVM), k-Nearest Neighbors (KNN), Multilayer Perceptron (MLP), XGBoost (XGB), LightGBM (LGBM), and Naive Bayes (NB). Metamodels were created to address overfitting and underfitting issues. The proposed system achieves a test accuracy of 91.45% and a train accuracy of 94.10%, with F1 scores of 0.91 and 0.94, respectively. The primary contributions of this study are threefold: introducing a summary-based legal outcome prediction framework, providing a systematic and multilingual evaluation of legal outcome prediction across English and regional language summaries, and demonstrating the effectiveness of stacked ensemble learning in enhancing predictive performance and robustness. By predicting outcomes from the legal summaries in regional languages, the study promotes inclusivity and trust in the legal system for regional language speakers. This work advances legal analytics and offers a practical solution for expanding access to justice in diverse linguistic contexts, setting a precedent for future research and applications in legal technologies. The code and dataset for the work is available at: https://github.com/ppati-git/Legal_CaseOutcome_Prediction.