A Text Mining Approach for Automated Case Classification of Judicial Judgment

Court judgments are based on legal reasoning, evidence, and judicial decisions. However, many judgments are difficult to understand due to their length and complex language. Judges and attorneys often cite legal rulings to interpret regulations effectively. Legal education for lawyers, judges, and trainees benefits from judicial insights into legal reasoning, legislative interpretation, and legal standards. Accurate classification of legal judgments requires sophisticated methodologies. The increasing number of new and pending cases adds to the courts' workload, highlighting the need for efficient classification. Since only a limited number of court rulings are decided each year, many cases remain unresolved and are carried forward to the following year. Despite efforts to improve classification accuracy across judicial judgment datasets, Pakistani court judgments still lack a proper classification system. To address this gap, we compiled a dataset of criminal judgments from the Pakistani High and Supreme Courts. We developed a machine learning model using neural networks and a transformer architecture to achieve accurate classification. Our approach incorporated Support Vector Machine (SVM), Long Short-Term Memory (LSTM), and DistilBERT models, taking into account the dataset’s volume and unique characteristics. Our research explores optimal classification algorithms tailored to Pakistan’s legal landscape. After comparison, the DistilBERT transformer model outperformed the others, achieving an accuracy of 98%. It demonstrated an exceptional ability to understand contextual semantics and effectively handle the complexities of multi-label classification in legal judgments. The contributions of our work include the development of a new dataset of Pakistani court rulings, advancements in legal research, and an automated case classification system designed to streamline judicial procedures and enhance access to justice.