基于因果推断和多头自注意力机制的学生成绩预测

Students′performance prediction based on causal inference and multi⁃head self⁃attention mechanism

学生成绩预测旨在为学校管理者提供决策支持,帮助教师改进教学手段,对学生进行学业指导,最终提高学生成绩。传统的学生成绩预测方法大多利用相关性分析选取重要因素,忽视多变量之间的间接联系和联系的方向性,而且在进行预测时没有区分输入特征的重要程度,因此提出一种基于因果推断和多头自注意力机制的学生成绩预测方法。该方法不仅使用因果推断选取与标签具有方向性联系的直接特征和间接特征,而且用多头自注意力机制区分不同特征对期末成绩的影响程度。在公开数据集上进行大量的实验,结果显示所提出方法的预测准确率高达93.06%,高于其他传统成绩预测方法。

The purpose of students′performance prediction is to provide decision support for school administrators,help teachers improve teaching methods,provide academic guidance to students,so as to ultimately improve the students′performance.In the traditional students′performance prediction methods,correlation analysis is mostly used to select important factors,and the indirect connection and directionality of the connection between multiple variables are ignored.Moreover,the degree of importance of input features is not distinguished when making predictions.Therefore,a students′performance prediction method based on causal inference and multi⁃head self⁃attention mechanism is proposed.In this method,causal inference is used to select the direct features and indirect features that have directional connection with labels,the multi⁃head self⁃attention mechanism is used to distinguish the influence degree of different features on the final results.A large number of experiments were carried out on an open data set.The experimental results show that the prediction accuracy rate of the proposed method can reach 93.06%,which is higher than the other traditional performance prediction methods.