Efficient Prediction of Judicial Case Decisions Based on State Space Modeling

With the rapid advancement of information technology and artificial intelligence, the digitization of legal texts has caused a swift increase in the volume of legal materials. Judges now face increased professional demands, larger information loads, and more complex case structures, which heightens their workload and demands. To enhance the quality and efficiency of judicial work and drive the modernization of the judicial system, the application of intelligent prediction models has become essential. This paper presents the MambaEffNet model, which integrates multiple modules such as Convolutional Neural Networks (CNN) and Multilayer Perceptrons (MLP). The core convolutional structure is improved using a state space model, and a multi-directional feature fusion structure is designed to enhance the performance of sequence feature extraction. Generative Adversarial Networks (GAN) are employed for data augmentation, to address the issue of missing features in judicial case predictions. The EfficientNetV2 architecture is used to optimize the kernel size and the expansion ratio of input and output channels. Experimental results demonstrate that the MambaEffNet model achieves a prediction accuracy of 92.05% on the Nigerian Supreme Court judgment dataset and performs excellently on other judicial datasets, significantly improving prediction accuracy and efficiency. Specifically, the MambaEffNet model increases the prediction accuracy for criminal and civil case judgments by 9.53% and 11.57%, respectively. Additionally, the model excels in handling long sequence data, effectively capturing key features and providing comprehensive decision support.