Prediction and Analysis of Judicial Small Data Based on Mechanism Model and System Identification

The development of legal artificial intelligence (AI) has strongly contributed to the fairness and efficiency of the law. Legal AI involves fundamental rights such as personal freedom and property, making the reliability of the calculation results critical. Specifically, for legal scenarios with a small data size, commonly used intelligent algorithms have clear limitations. First, for traditional statistical methods, when analyzing legal scenarios with a small data size, it is difficult to directly apply asymptotic estimation methods applicable to big data. Thus, it is difficult to ensure the accuracy of parameter estimates and the reliability of calculations. Second, both deep-learning and large-language-model methods require a sufficiently large data size, which leads to limited recognition and learning capabilities for specific scenarios with a small data size. Based on the clarity of laws and the certainty of legal operations, we propose an approach based on mechanism models and system identification. By incorporating priori system structures and elements in a computational model, we can effectively improve the prediction accuracy of small-data scenarios. To test the practical effect of small-data interpretable sentencing research in a law application, we applied the mechanism model and system-identification algorithm to actual judicial practice. The calculations for specific judicial-verdict data showed the high accuracy of the sentence prediction, confirming the reliability of the above model and calculation method, as well as reflecting the superiority of the nonlinear sentencing model and the effectiveness of nonlinear recursive-identification theory.