基于异构距离的集成分类算法研究

PDF下载HTML

Title:

Imbalanced heterogeneous data ensemble classification based on HVDM-KNN

作者:

张燕, 杜红乐

商洛学院 数学与计算机应用学院, 陕西 商洛 726000

Author(s):

ZHANG Yan, DU Hongle

School of Math and Computer Application, Shangluo University, Shangluo 726000, China

关键词:

异构数据; 不均衡数据; 异构距离; 集成学习; 过取样; 欠取样

Keywords:

heterogeneous data; imbalanced data; heterogeneous value difference metric; ensemble learning; over sampling; undersampling

分类号:

TP391.4

DOI:

10.11992/tis.201807023

摘要:

针对异构数据集下的不均衡分类问题，从数据集重采样、集成学习算法和构建弱分类器3个角度出发，提出一种针对异构不均衡数据集的分类方法——HVDM-Adaboost-KNN算法（heterogeneous value difference metric-Adaboost-KNN），该算法首先通过聚类算法对数据集进行均衡处理，获得多个均衡的数据子集，并构建多个子分类器，采用异构距离计算异构数据集中2个样本之间的距离，提高KNN算法的分类准性能，然后用Adaboost算法进行迭代获得最终分类器。用8组UCI数据集来评估算法在不均衡数据集下的分类性能，Adaboost实验结果表明，相比Adaboost等算法，F₁值、AUC、G-mean等指标在异构不均衡数据集上的分类性能都有相应的提高。

Abstract:

A novel classification method, the heterogeneous value difference metric-Adaboost-KNN (HVDM-Adaboost-KNN), is proposed to achieve data resampling, to obtain an ensemble learning algorithm, and to construct a weak classifier for addressing the imbalanced classification of a heterogeneous dataset. This algorithm initially equalizes the dataset using a clustering algorithm to obtain several equalized data subsets and constructs several sub-classifiers. Further, the heterogeneous distance is used to calculate the distance between two samples in the heterogeneous dataset to improve the classification accuracy of the KNN algorithm. Subsequently, the Adaboost algorithm is used to iteratively obtain the final classifier. Eight groups of UCI datasets are used to evaluate the classification performance of the algorithm in imbalanced datasets. The Adaboost experimental results denote that the classification performance of indices, such as the F1 value, AUC, and G-means, using the heterogeneous imbalanced datasets was better when compared with that exhibited by other algorithms.