Multimodule Legal Analytics Pipeline for Outcome Prediction and Transparent Justification in Commercial Cases
Indian commercial courts face heavy caseloads, increasing the need for rapid legal research and reliable case outcome prediction. To address this, an AI-powered legal research engine was developed, tailored for Indian commercial litigation under the Commercial Courts Act, 2015. The system processes statutes and judgments to build a semantic search index and uses an LSTM-based neural model to predict outcomes (petitioner win or loss). Training is done on the ILDC dataset containing annotated court judgments. For interpretability, a LIME-based module highlights critical facts and legal provisions. A fine-tuned LLaMA language model then generates human-readable justifications aligned with judicial reasoning by leveraging LSTM outputs and LIME attributions. The model achieves 91% accuracy and 0.95 AUC, with a balanced confusion matrix. The explainability layer emphasizes critical contract terms and legal citations, enhancing transparency. This system also addresses limitations such as data bias, domain shifts in lower courts, and ethical concerns. It aims to streamline research and improve legal decision support.
Paper
Full text
Multimodule Legal Analytics Pipeline for Outcome Prediction and Transparent Justification in Commercial Cases
Semantic Scholar · 2025
Abstract
Indian commercial courts face heavy caseloads, increasing the need for rapid legal research and reliable case outcome prediction. To address this, an AI-powered legal research engine was developed, tailored for Indian commercial litigation under the Commercial Courts Act, 2015. The system processes statutes and judgments to build a semantic search index and uses an LSTM-based neural model to predict outcomes (petitioner win or loss). Training is done on the ILDC dataset containing annotated court judgments. For interpretability, a LIME-based module highlights critical facts and legal provisions. A fine-tuned LLaMA language model then generates human-readable justifications aligned with judicial reasoning by leveraging LSTM outputs and LIME attributions. The model achieves 91% accuracy and 0.95 AUC, with a balanced confusion matrix. The explainability layer emphasizes critical contract terms and legal citations, enhancing transparency. This system also addresses limitations such as data bias, domain shifts in lower courts, and ethical concerns. It aims to streamline research and improve legal decision support.