半监督学习在不平衡样本集分类中的应用研究

在对不平衡样本集进行分类时容易产生少数类样误差大的问题,而目前半监督学习中的算法多数是针对未有明显此类特征的数据集。针对一种半监督协同分类算法在该问题上的有效性进行了研究。由于进一步增强了分类器差异性,该算法在理论上对不平衡样本集具有良好的分类性能。根据该算法建立分类模型,利用其对桥梁结构健康数据进行分类实验,与Tri-Training算法的结果比较表明,该算法对不平衡样本集具有良好的适用性,从而验证了上述算法的有效性。

Semi-supervised learning in imbalanced sample set classification

Higher error rate emerged in the minority class of samples when make classification on imbalanced sample set, but most algorithms in semi-supervised learning are based on normal data set. This paper studied the effectiveness of a semi-supervised collaboration classification method. Because of the further enhanced classifier difference, this algorithm had good performance on classification of imbalanced sample set. It established classification model based on the above algorithm, and used this model to make classification with bridge structural health monitoring data, the compared results of which demonstrated the applicability to imbalanced sample set. Therefore it validated the effectiveness of the algorithm.