基于特征重要度和决策树技术的考试成绩分析研究-工程硕士专业论文.docxVIP

I I 摘 要 摘 要 利用机器学习与数据挖掘方法对学生的考试成绩进行分析并从中发掘有利于教 学的可用知识是当前计算机辅助教学的重要组成部分。现有的成绩分析研究主要是选 用特定的数据挖掘方法来对考试成绩数据进行分类。这种处理方式忽略了考试成绩数 据本身所隐含的信息，例如，不同课程对于学生综合评价的影响等。所以需要弥补现 有考试成绩分析研究的不足。 本文基于某校开办的 10 类不同成人教育培训的考试成绩，主要进行了以下三个 方面的分析与研究：首先，采用基于互信息技术的特征重要度计算方法来对考试成绩 数据进行预处理，从而确定出与综合评价最为相关的教学课程，为优化教学课程的时 间安排提供科学依据；之后，利用决策树方法来对预处理之后的考试成绩进行数据挖 掘，通过对比不同决策树的学习结果，抽取标识课程学习与综合评价关系的学习规则， 方便考生依据抽取规则确定自己的学习规划；最后，执行了详细地实验验证，比较了 三种决策树方法在原始数据集以及特征选择之后数据集上面的分类表现，实验的结果 表明本文提出的考试成绩分析策略是可行和有效的，对于考生学习计划的制定以及考 试成绩的提高具有一定的参考作用。 关键词 考试成绩分析 互信息 特征选择 决策树 机器学习 数据挖掘 II II Abstract Abstract It is a very important part that using machine learning and data mining methods to analyze the examination score and further extract the useful knowledge. The current studies on examination score analysis can be regarded as the design to a supervised learning system, that is to say, classifying the examination score data with the specific data mining methods. The main limitation of the current studies is that these methods all neglect the underlying information hidden in the examination score data, e.g., the dependence between different courses and the impact of this dependence on the comprehensive evaluation. So we need to improve the defect of existing examination score analysis methods, This thesis presents the following three main contributions based on the examination scores deriving from the 10 different adult education projects. Firstly, preprocess the examination score data by using the feature selection based on mutual information technology so as to determine the maximally dependent courses with comprehensive evaluation and meanwhile reduce the teaching time of weak correlation. Secondly, conduct the data mining based on the preprocessed datasets and extract the related learning rules in order that the students can make the learning project as convenient as possible. Finally, carry out the detailed experiments to compare the performance of three different decision tree algorithms. The experimental results show