【摘要】 化学样本数据常为非平衡,用传统方法分析这些数据集时,对于需特别关注的少数类数据,识别能力往往较差。因此,提出建立基于粒计算的分类规则模型(GCCRM),先用改进的自适应共振网络ETM-ART2将性质相近的个体聚合为信息粒,降低样本容量和问题规模,同时又保持较高的纯度;然后将信息粒的属性特征模糊离散化,简化它;最后经关联规则挖掘,得可预测样本的分类规则模型。用于识别玻璃,结果GCCRM能剔除冗余信息,加强关键特征,所提取的分类规则预测正确率高,尤适用于非平衡数据集,其性能比C4.5决策树、支持向量机SVM等算法优良。更多还原

【Abstract】 Chemical data is often imbalanced,when analyzes these data,traditional machine learning algorithms usually has poor recognition capability on the concerned minority class.Therefore,a granular computing based classification rules mining model(GCCRM) is proposed.ETM-ART2 is utilized to construct information granulations(IGs),then each IGs’ attribute is fuzzy transformed to three linguistic terms,which can simplify research objects.Association rules mined from these IGs are applied to decision for unknown patterns. Then GCCRM is utilized to glasses recognition problem,and the experimental results comparing with C4.5 and SVM demonstrate the superiority of this method,especially for imbalanced data.更多还原