基于K折交叉验证的选择性集成分类算法

K-Fold Cross-Validation Based Selected Ensemble Classification Algorithm

针对传统选择性集成方法确定个体分类器权重参数不准确、计算复杂度较高的不足,提出了一种基于K折交叉验证的选择性集成分类算法。该算法首先采用集成学习思想训练一定数目的分类器,然后对每一个分类器设定权重参数初值,并利用交叉验证思想确定对应最大平均分类准确率的参数作为最终的个体分类器的权重因子,最后将权重小于某个预设阈值的分类器剔除,完成选择性集成学习。由于交叉验证方法可以较快并且较为精确地进行权重参数的确定,所以本算法可以有效地提高选择性集成方法的分类性能。在UCI标准数据集上的仿真实验充分证明了本算法的有效性。

As traditional selected ensemble method can not determine individual classifier weight parameters accurately, and has a high computational complexity, this paper proposes a K-fold cross-validation based selected ensemble classification algorithm. The algorithm firstly uses ensemble learning strategy to train a certain number of classifiers, and then sets the fight initial value of the weight parameter for each classifier, and use cross-validation strategy to determine the fight weight parameters corresponding to the maximum average classification precision. Finally, remove the classifiers whose weight parameter is less than a preset threshold to complete the ensemble learning. As cross-validation method can determine the weight parameters faster and more accurately, so the proposed algorithm can effectively improve the classification performance of the selected ensemble method. The simulation experiments on the UCI standard data sets fully prove the effectiveness of the proposed algorithm.