Abstract: Concerning the relevance between the attributes and the contribution difference of attribute values to attribute weights in classification dataset, an improved classification model and the formulas for calculating the impact factor and sample forecast information were proposed based on mutual information. And the classification model predicted the unlabelled object classes with the sample forecast information. Finally, the experimental results show that the classification model based on mutual information can effectively improve forecast precision and accuracy performance of classification algorithm.

Key words: mutual information, average mutual information, classification model, impact factor, sample forecast information

摘要： 针对分类数据集中属性之间的相关性及每个属性取值对属性权值的贡献程度的差别，提出基于互信息量的分类模型以及影响因子与样本预测信息量的计算公式，并利用样本预测信息量预测分类标号。经实验证明，基于互信息量的分类模型可以有效地提高分类算法的预测精度和准确率。

关键词: 互信息量, 平均互信息量, 分类模型, 影响因子, 样本预测信息