基于改进的Stacking模型的教师分数预测
Teacher score prediction based on improved Stacking models
杨雪亭 1赵霞 1王钊1
作者信息
- 1. 河北经贸大学管理科学与信息工程学院,河北石家庄 050061
- 折叠
摘要
针对单个模型泛化能力弱的问题,提出了基于改进的Stacking模型的教师分数预测模型.首先用LightGBM进行特征选择,再对目标值进行Box-Cox转换,最后利用基于Stacking的回归模型,集成了 AdaBoost(自适应增强)、RF(随机森林)、XGBoost(极限梯度提升)、KRR(核岭回归)四种算法,再用Stacking模型与单个学习器Ridge(岭回归)加权组合预测教师分数.结果显示,在教师分数数据集上改进后的算法均方根误差为9.715,较回归算法AdaBoost降低了 0.227,较传统Stacking融合模型降低了 0.161,该模型有效地提高了预测准确率.
Abstract
This paper proposes an improved Stacking-based teacher score prediction model for the problem of weak generalisation ability of a single model,firstly using LightGBM for feature selection,then Box-Cox transformation for target values,and finally using Stacking-based regression model that integrates AdaBoost,RF,XGBoost,KRR algorithms,and then the Stacking model is weighted with a single learner Ridge combination to predict teacher scores.Aiming at the problem of weak generalization ability of a single model,a teacher score prediction model based on improved Stacking model is proposed.The experimental results show that the root mean square error of the improved algorithm on the teacher scores dataset is 9.715,which is 0.227 lower than the regression algorithm AdaBoost and 0.161 lower than the traditional Stacking fusion model.
关键词
教育数据挖掘/学生评教/Stacking集成学习/特征选择Key words
Educational data mining/Student assessment of teaching/Stacking integrated learning/Feature selection引用本文复制引用
基金项目
河北省新工科研究与实践项目(第二批)(2020GJXGK026)
出版年
2024
NETL
NSTL
万方数据