针对增加集成学习Bagging算法中分类器的差异性, 提高集成学习算法模型的鲁棒性, 研究了基于k-means聚类技术对集成学习算法Bagging进行剪枝。在基础Bagging算法中融合对Bagging分类器的聚类, 然后在不同簇中选择具有代表价值的分类器为最终集成学习预测结果投票, 并在多个机器学习数据集上验证这种提高差异性的方法与基本Bagging性能的差异。经过仿真实验最终得出在算法迭代10次的前提下, 改进的Bagging算法较常规Bagging算法在10个实验数据集中提高了7个数据集的预测精度, 其精度提高的平均值在3%; 在算法迭代100次的前提下, 改进的Bagging算法较常规Bagging算法在10个实验数据集中提高了9个数据集的预测精度, 其精度提高的平均值为2.5%。为复杂数据库环境下Bagging算法的应用提供了新思路。

Aiming at increasing the diversity of classifiers in the ensemble learning Bagging algorithm and improving the robustness of the ensemble learning algorithm model, this paper studies the pruning of Bagging, an ensemble learning algorithm, based on k-means clustering technology. In the basic Bagging algorithm, the clustering of Bagging classifiers is fused, and then the representative classifiers are selected in different clusters to vote for the final ensemble learning prediction results, and the difference of this method in improving the performance of basic Bagging is verified on multiple machine learning datasets. The simulation results show that the improved Bagging algorithm improves the prediction accuracy of 7 datasets in 10 experimental datasets by an average of 3% compared with the conventional Bagging algorithm on the premise of 10 iterations, and the improved Bagging algorithm is better than the conventional Bagging algorithm on the premise of 100 iterations. The prediction accuracy of 9 data sets is improved in 10 experimental data sets, and the average improvement of the accuracy is 2.5%. It provides a new idea for the application of Bagging algorithm in complex database environment.