摘要
目的:旨在对比学生成绩预测模型。方法:共获取605条数据,共32个解释变量,通过特征选择剩余23个自变量,建立训练集和测试集,以G3为目标变量,分别构建SVM模型、神经网络模型和逐步回归模型,比较这三个模型的预测结果。结果:采用逐步回归模型预测学生成绩在最小误差、最大误差、平均误差、平均绝对误差和标准差方面均低于神经网络模型和SVM模型,在预测值和实际值的线性相关系数方面均高于神经网络模型和SVM模型。结论:在预测学生成绩上,逐步回归模型优于神经网络模型和SVM模型。
Objective To compare students’achievement prediction models.Methods A total of 605 pieces of data and 32 explanatory variables were obtained.The remaining 23 independent variables were selected through characteristics to establish the training set and test set.Taking G3 as the target variable,SVM model,neural network model and stepwise regression model were constructed respective⁃ly to compare the prediction results of the three models.Result The results of stepwise regression model were lower than neural network model and SVM model in minimum error,maximum error,mean error,mean absolute error and standard deviation,and higher than neu⁃ral network model and SVM model in linear correlation coefficient of predicted value and actual value.Conclusion Stepwise regression model is better than neural network model and SVM model in predicting student achievement.
作者
王欣欣
汤军
WANG Xin-xin;TANG Jun(Yangtze University,School of Earth Sciences,Wuhan 430100,China)
出处
《电脑知识与技术》
2020年第1期199-202,共4页
Computer Knowledge and Technology
基金
顾及空间上下文关系的能量最小化地图多要素协同移位方法(国家自然科学基金青年基金41701537)
关键词
学生成绩预测模型
神经网络模型
逐步回归模型
SVM模型
线性相关系数
Students’Achievement Prediction Model
Neural Network Model
Stepwise Regression Model
SVM Model
Linear Correla⁃tion Coefficient