【摘要】 关于食品安全谣言的文本数据是极其典型的不平衡数据集之一，而传统机器学习算法对于不平衡数据少数类的分类精度较低。动态选择是评估分类器池中每个分类器对测试样本局部区域进行预测的能力，根据对于预测能力的评估，为每个测试样本选择分类器子集的方法。提出一种针对多类不平衡数据集的动态选择方法DCS-MI,并进行了广泛的实验。与一些最先进的动态选择技术相比，该方法提高了对不平衡数据集的分类性能，并能够在食品领域谣言分类问题上得到很好的应用。更多还原

【Abstract】 Text data of food safety rumors is a kind of very typical imbalanced datasets. The traditional machine learning methods have a low classification accuracy for minority classes of imbalanced datasets. Dynamic selection is a method to evaluate the ability of each classifier in the classifier pool to predict the local region of test instances, and select a subset of classifiers for each test instance according to the evaluation of prediction ability. This paper proposes a new dynamic selection method named DCS-MI for multi-class imbalanced datasets. Extensive experiments were carried out. Compared with some of the state-of-the-art dynamic selection technologies, it improves the classification performance on imbalanced datasets, and can be well applied to the rumor classification in the food field.更多还原