集成分类对比:Bagging NB & Boosting NB

Integrated Classification Comparison:Bagging NB & Boosting NB

Bagging和Boosting是两种重要的投票分类算法,前者并行生成多个分类器,后者通过调整样本权重,串行生成多个分类器.将Bagging与Boosting算法与朴素贝叶斯算法相集成,构建了Bagging NB和AdaBoosting NB算法.以UCI数据集为基础,进行实验对比,结果表明,Bagging NB算法较为稳定,可以产生优于NB算法的分类结果,而Boosting算法受到数据分布中的奇异值影响较大,部分数据集上与NB算法的基础效果较差.

Bagging and Boosting are two important voting classification algorithms. Bagging parallel generates multiple classifiers and by adjusting the sample weights, Boosting generates multiple classifiers serially. This paper integrated Bagging and Boosting algorithms with Naive Bayesian to construct Bagging NB and Adat3oosting NB. Based on UCI data sets, experiment result show that Bagging NB is more stable and can produce more accurate classifier than NB. Boosting algorithms are sensitive with the distribution of the data set and sometimes less effective.