冯照石　范祺

摘 要：针对朴素贝叶斯算法在处理不平衡数据时准确率不高的问题，提出类权重和属性值权重相结合的双重加权朴素贝叶斯算法.双重加权可降低算法的属性独立假设对分类结果的影响，提升少数类对分类过程的作用.乳腺癌患者复发率预测结果表明，双重加权朴素贝叶斯算法相对于传统的朴素贝叶斯算法、属性值加权的朴素贝叶斯、K最近邻分类算法、支持向量机分类算法、随机森林算法，准确率上分别从0.72，0.79，0.77，0.80，0.81提升至0.84，精确率和召回率均有不同程度的提升.

关键词：算法改进;数据不平衡;双重加权;乳腺癌复发

[中图分类号]O171   [文献标志码]A

Double Weighted Naive Bayes Algorithm for PredictingRecurrence Rate of Breast Cancer

FENG Zhaoshi，FAN Qi*

（School of Computer Science and Technology，Huaibei Normal University，Huaibei 235000，China）

Abstract：Aiming at the problem of low accuracy of Naive Bayes algorithm in dealing with unbalanced data，a double weighted Naive Bayes algorithm is proposed，which combines class weight and attribute value weight.The double weighting reduces the influence of attribute independence hypothesis on classification results and improves the effect of a few classes on classification process.The experimental results show that compared with the traditional Naive Bayes algorithm，Naive Bayes algorithm with weighted attribute values，K nearest neighbor classification algorithm，Support Vector Machine classification algorithm and Random Forest algorithm，the accuracy of the double weighted Naive Bayes algorithm is increased from 0.72，0.79，0.77，0.80 and 0.81 to 0.84 respectively，and the accuracy and recall rate are also improved to varying degrees.

Key words：improved algorithm;unbalanced data;double weighting;recurrence of breast cancer

樸素贝叶斯算法是分类算法中的经典算法之一，分类算法简单且十分高效.算法本身是基于独立性假设条件，且在面对不平衡数据时分类结果往往有所降低，故人们在提高朴素贝叶斯的分类性问题上做了广泛的研究和改进.邓维斌[1]等提出了一种基于粗糙集的属性加权方法.Hall[2]提出了一种基于决策树的属性加权方法.Chao Qun[3]等将一些关于朴素叶斯的工作应用到VDM的重量训练实例中，分配根据估计的条件概率与真正的条件概率的区别，对体重进行反复调整.Jiang[4]等提出根据不同类中属性分配特定权重的细粒度加权方法，使模型在分类性能上具有一定优势.

肿瘤阳性预测问题中，数据具有严重的不平衡性，如何提高分类结果的准确性成为数据挖掘中的重要研究方向，具有重要……