An Imbalanced Data Robust Classification Model for Student Achievement Prediction

In this study, the mining and analysis of student scores are used to explore the impact of student behaviors on their final performance and provide guidance for teaching by predicting student achievement. Traditional performance prediction methods often ignore the problem of imbalanced classes and noise samples, which makes the model overfit most classes. In this paper, we propose an imbalanced data classification model for student achievement prediction. First, a synthetic minority oversampling technology based on anomaly tolerance K-means clustering is used to expand the minority sample to optimize the sample set. Second, the cluster model is utilized to predict students’ grades. Comparative experiments have shown that our model improves the prediction accuracy, which could reach 99.32% and 99.44% compared with other machine learning methods on two open datasets.