集成方法在极限学习机中的应用

Application of ensemble methods in extreme learning machine

针对传统的单一机器学习模型对非平衡数据集分类预测性能偏低的问题,通过用Adaboost策略将传统的最大化Gm代价调整极限学习机集成起来,生成一种最大化Gm集成极限学习机分类模型(MG-CCR-EELM),使其能够适用于不同平衡率非平衡数据集的分类。通过与现有的最大化Gm代价调整极限学习机、代价敏感混合属性多决策树、改进的模糊支持向量机、随机森林等用于非平衡数据集的分类模型的实验对比,MG-CCR-EELM模型在UCI公共数据集上的准确率最高可提升3.01%,在经颅多普勒数据集上的分类预测准确率提升了5.67%,验证了MG-CCR-EELM模型是一种有效的集成学习模型。

In response to the problem of low classification and prediction performance of traditional single machine learning algorithm models for imbalanced datasets,an adaptive boosting strategy is used to integrate the traditional maximum Gm cost adjustment extreme learning machine,generating a Maximizing Gm Class⁃specific Cost Regulation Ensemble Extreme Learning Machine(MG-CCR-EELM),which can be suitable for classification of imbalanced data with different features.Compared with the existing maximizing Gm Class⁃specific Cost Regulation Extreme Learning Machine,cost sensitive mixed attribute multiple decision tree,improved fuzzy support vector machine,random forest and other classification models used for unbalanced data sets,the accuracy of MG-CCR-EELM model on UCI public data set can be improved by up to 3.01%,and the accuracy of classification prediction on Doppler extracranial artery Transcranial Doppler data set can be improved by 5.67%,the MG-CCR-EELM model has been validated as an effective ensemble learning model.