基于MTS-AdaBoost的不平衡数据分类研究

Classification of unbalanced data based on MTS-AdaBoost

不平衡数据在实际应用中广泛存在,而传统的分类算法大多假定类分布平衡,因此解决不平衡数据的分类问题已经成为数据挖掘的瓶颈问题之一。马田系统(MTS)是一种多元模式识别方法,将其与Ada Boost集成算法相结合,形成MTS-Ada Boost算法。该算法以MTS为基分类器,根据上一个基分类器的预测结果,自行调整下一个基分类器中样本被抽中的概率,以此来改变不同类数据的平衡度。最后,利用该算法对2010—2015年间上市公司的财务危机预警进行实证研究,结果表明,MTS-Ada Boost算法在系统降维和分类效果上都优于传统MTS,也优于其他常用的单一分类器。

Unbalanced data are widely used in practical applications, but most of the traditional classification algorithms as- sume class distribution balance. Therefore, solving the problem of unbalanced data classification has become one of the bottle- necks in data mining. MTS is a multivariate pattern recognition method, which is combined with the AdaBoost integration algo- rithm to form the MTS-AdaBoost algorithm. The algorithm used the MTS as the base classifier, and adjusted the probability of the sample in the next base classifier according to the prediction result of the previous base classifier, so as to change the ba- lance degree of the different class data. Finally, this paper applied this method to research the financial crisis warning of listed companies from 2010 to 2015. The result shows that MTS-AdaBoost algorithm＇ s dimensionality reduction and classification resuits are both superior to traditional MTS, and they are also superior to other commonly used single classifiers.