Research on Performance Prediction Model Based on Innovative Methods

In the field of education, academic performance is an important indicator for evaluating student learning outcomes. This paper proposes a student performance prediction model based on a dual-path attention mechanism, which predicts students' future performance by analyzing their personal information and behavioral data. The paper first conducts a detailed feature construction and evaluation of the dataset, then uses various traditional machine learning algorithms for performance prediction, and proposes a feature evaluation method based on Grad-CAM. The experimental results show that the TWA model performs excellently in the performance prediction task, achieving a prediction accuracy of 72.43, significantly outperforming traditional machine learning algorithms. This research provides valuable insights for personalized analysis and teaching improvement in the field of education. Practically, TWA enables educators to allocate teaching resources more effectively by highlighting the key factors influencing each student's performance. It also allows for the creation of more targeted learning materials, enhancing the efficiency of the overall teaching-learning process.