应用朴素贝叶斯算法预测学生学习指标的改进研究  

Study on Improvement of Using Naive Bayes Algorithm to Forecast Student's Learning Indicator

作　　者：李宏林 

机构地区：[1]泉州医学高等专科学校,福建泉州362000

出　　处：《情报探索》2015年第6期22-26,32,共6页Information Research

基　　金：福建省教育厅B类科技项目"基于数据挖掘的学生成绩分析与研究"(项日编号:JB12317)成果

摘　　要：以综合因子分值作为学生学习指标,应用SVM、KNN、决策树、迭代森林、朴素贝叶斯、神经网络六种数据挖掘算法对其进行预测,从合并样本随机拆分及人工指定划分两个角度,根据学生科目成绩预测学生学习指标发展趋势,发现朴素贝叶斯算法预测准确率最高。为提高预测准确率,提出朴素贝叶斯算法的3种改进途径,即主成分分析、先验概率修正、多重过滤预判;利用JAVA软件建立一个基于PMML的素质预测系统,采用主成分分析、贝叶斯过滤、贝叶斯先验概率修正3个模块来提高对综合学习素质的预测准确率。The paper takes comprehensive factor score as student’s learning indicator, uses 6 data mining algorithms of SVM, KNN, decision tree, iterative forest, Naive Bayes and neural network to forecast it. The paper finds out that the accuracy of Naive Bayes is highest in forecasting development trend of student’s learning indicator on the basis of their subject achievements from two perspec-tives of pooled sample randomly splitting and artificially dividing. In order to improve forecast accuracy, the paper puts forward 3 im-provement ways for Naive Bayes algorithm, including principal component analysis （PCA）, prior probability correction and multiple filter anticipation. The paper uses JAVA software to build a PMML-based quality forecast system, which adopts 3 modules of principal com-ponent analysis, Bayesian filtering and Bayesian prior probability correction to improve accuracy of comprehensive learning quality forecast.

关 键 词：数据挖掘 预测分析 朴素贝叶斯 算法改进 PMML 

分 类 号：C40-058[社会学]