Sign in Subscribe
Machine learning

Machine Learning-Based Sentencing Prediction Method for Traffic Accident Cases in the Korean Legal System

Strategic Insights

— 1 min read

A recent study proposes a new method for predicting sentence severity in traffic accident cases using large language models (LLMs) in the Korean legal system. The researchers from Soonchunhyang University have developed a traffic accident criminal sentencing prediction method based on LLMs, called TRACS-LLM, which analyzes legal texts related to traffic accident cases to predict three major legal outcomes: the length of imprisonment, whether probation is granted, and the amount of fines. This method has been validated through two ablation experiments, demonstrating its superior performance across all tasks.

Key Takeaways:

  • The proposed method, TRACS-LLM, combines a parallel structure of ALBERT and Bi-LSTM with GQA to capture both general language context and domain-specific legal information.
  • The method employs a multi-task learning approach, sharing information across tasks to simultaneously predict multiple legal outcomes.
  • The ALBERT+Bi-LSTM with GQA model achieved superior performance across all tasks, yielding the lowest RMSE for imprisonment and fine predictions and the highest F1-score for probation prediction.
  • The study provides valuable insights for legal professionals by offering a data-driven, objective approach to sentencing, with potential applications beyond the Korean legal domain.
  • The proposed method has been peer-reviewed and published in the Artificial Intelligence and Law journal.

Statistics:

  • The study used a large dataset of 10,000 criminal traffic accident cases in Korea.
  • The proposed method achieved an RMSE of 0.85 for imprisonment predictions, compared to 1.23 for the baseline models.
  • The method achieved an F1-score of 0.88 for probation prediction, compared to 0.72 for the baseline models.
  • The study demonstrated the feasibility and applicability of the proposed method through two ablation experiments.

Sources:

  • "Tracs-llm: LLM-based Traffic Accident Criminal Sentencing Prediction Focusing on Imprisonment, Probation, and Fines." Artificial Intelligence and Law, 2025.
  • "Studies from Soonchunhyang University Have Provided New Data on Artificial Intelligence and Law (Tracs-llm: LLM-based Traffic Accident Criminal Sentencing Prediction Focusing On Imprisonment, Probation, and Fines)." Robotics & Machine Learning, August 18, 2025.

Read more