基于K-XGBoost融合模型的高校学生学情预测研究  

Research on Undergraduate Academic Prediction Based on K-XGBoost Fusion Model

作　　者：赵宇奔 王鑫宁 李崇[1] Zhao Yuben;Wang Xingning;Li Chong(College of Engineering,Ocean University of China,Qingdao 266100,China;Teaching Center of Fundamental Courses,Ocean University of China,Qingdao 266100,China)

机构地区：[1]中国海洋大学工程学院,山东青岛266100 [2]中国海洋大学基础教学中心,山东青岛266100

出　　处：《南京师大学报（自然科学版）》2023年第3期89-97,共9页Journal of Nanjing Normal University(Natural Science Edition)

基　　金：中央高校基本科研业务费专项(202213016);山东省自然科学基金项目(ZR201910230031);2022年度青岛市社会科学规划研究项目(QDSKL2201014).

摘　　要：高精准的学情预测是提升高校教学水平促进教学改革的重要技术手段.目前学情预测存在数据维度单一和数据结构不平衡等问题,降低了预测模型的准确性与泛化能力.为此,本文提出了K-XGBoost学情预测融合模型.首先,该模型通过精准特征提取与重构,构建基于高校教务处数据库的多维度学情特征集;其次,设计基于最小2-范数的聚类算法,创新性地建立无监督数据平衡化机制;最后,基于损失函数优化的XGBoost集成学习方法设计学情预测模块,构建高准确性和高泛化能力的K-XGBoost学情预测融合算法.实验结果表明,K-XGBoost多个子类模型的预测值均较好地逼近真实值,可将成绩预测结果的平均绝对误差(MAE)和均方根误差(RMSE)相较基线XGBoost模型分别降低了76.19%、85.33%,显著提升了学情预测的准确性和泛化能力.High-precision prediction of academic conditions is an important technical means to improve the teaching level of colleges and promote teaching reform.At present,there are problems such as single data dimension and unbalanced data structure in academic prediction,which reduces the accuracy and generalization ability of the prediction model.To the end,this paper proposes a K-XGBoost academic situation prediction fusion model.Firstly,through accurate feature extraction and reconstruction,the model constructs a multi-dimensional set of academic features based on the database of the Academic Affairs Office of the University.Secondly,the clustering algorithm based on the minimum 2-norm is designed,and the unsupervised data balancing mechanism is innovatively established.Finally,the XGBoost integrated learning method based on loss function optimization designs the academic situation prediction module,and constructs a K-XGBoost learning situation prediction fusion algorithm with high accuracy and high generalization ability.The experimental results show that the predicted values of K-XGBoost models can well approximate the real values,and the MAE and RMSE of performance prediction results are reduced by 76.19%and 85.33%respectively compared with XGBoost models,which significantly improves the accuracy and generalization ability of the academic performance prediction.

关 键 词：K-XGBoost 学情预测 数据挖掘 机器学习 集成学习 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]