Abstract: With the continuous development of information technology, extracting useful information from massive student data and predicting academic performance has become one of the hot topics among researchers. In order to predict student grades more accurately, this study employed the RFE_RF feature selection method to reduce the dimensionality of the student dataset. Additionally, an information entropy based weighted voting method was used to integrate the predictions from various classifiers. Moreover, Bayesian optimization algorithm was used to adjust the parameters of decision tree, random forest, logistic regression and naive Bayes to improve the performance of the student achievement prediction model. After experimentation and evaluation, the overall accuracy of the model reached 85.41%. Furthermore, the model exhibited good performance in terms of accuracy, F1-score, and recall rate across different grade levels and metrics.