一种不平衡水声目标数据的选择性集成算法

Selective ensemble algorithm for imbalanced underwater acoustic target data

针对不平衡水声目标数据分类问题,本文提出了一种间隔和差异性融合的选择性集成算法。从理论上给出了单纯增加差异性无法改善泛化性的原因,融合间隔和差异性构造了选择性度量,利用选择性度量对基分类器进行选择性集成从而形成最终分类器。实测水声目标数据试验结果表明:本文算法整体性能优于AdaBoost算法和常规选择性集成算法,说明其更适合处理不平衡水声目标数据分类问题。

To solve the problem of imbalanced underwater acoustic target data classification,in this paper,we propose a margin and diversity fusion selective ensemble algorithm(MDSE algorithm).First,we provide a theoretical explanation of why the generalization could not be improved by simply increasing diversity.Second,we develop a selective measurement technique that involves margin and diversity fusion.Finally,we obtain the final classifier from a selective ensemble of base classifiers using selective measurement.Underwater acoustic target data was obtained.The experimental results showed that the MDSE algorithm performed better than the AdaBoost and common selective ensemble algorithms,which means that the MDSE algorithm is more suitable for imbalanced underwater acoustic target data classification.