High school student GPA prediction by various linear regression models

Academic performance (GPA) is a significant index for high school students in North America. The research aims to develop and validate the best predictive regression model to evaluate the GPA of high school students. In addition, the prediction numeric data of GPA can correspond to specific classification data of GPA (Grade Level) to calculate and compare the models accuracy for predicting Grade Level. The research subjects are high school students from different backgrounds in North America, including their background, study habits, participation in extracurricular activities, etc. The experiment explores the impact of different factors on student GPA and finds that the number of absences from lectures is a key factor in predicting student GPA. Multiple linear regression analysis is used as the main model in the experiment, which may be improved by the stepwise regression methods. The generalization ability of the model is evaluated through cross-validation (CV) methods. Also, the boosting or random forest model is used to be the comparing model for predicting GPA. The experimental result shows that the multiple linear regression model has high accuracy (84%) and reliability (R^2 value is 0.95) in predicting student GPA. The conclusion of the research emphasizes the importance of predicting student GPA in high school education and the potential for guiding educational practice through data analysis. Future work will consider introducing more subjects and variables, such as different subject learning abilities, mental health, and social support, to further improve the predictive accuracy of the model.