WSS-Linear: A Transformer-Enhanced Approach for Exam Score Prediction

Exam scores play a crucial role in students' academic development and future planning; therefore, accurately predicting students' exam performance has become an important research topic in the field of education. Traditional exam score prediction methods predominantly use linear regression, which, while capable of achieving satisfactory predictive results, exhibits significant shortcomings in capturing nonlinear relationships between features and in identifying the importance of different features. To address this issue, this paper proposes a novel prediction method called WSS-Linear. This method introduces a transformer into traditional linear regression, effectively compensating for the deficiencies in capturing nonlinear relationships and enabling dynamic allocation of feature weights, thereby enhancing the model's ability to learn feature importance. Experimental results on the publicly available Kaggle dataset "Student Performance Factors" demonstrate that the WSS-Linear method outperforms traditional methods across multiple evaluation metrics, validating its effectiveness and reliability in exam score prediction.