基于平衡度调整策略的Ba-AL主动学习算法

Ba-AL Active Learning Algorithm Based on Balance Adjustment Strategy Active Learning Algorithm based on Balance Adjustment Strategy

提出一种面向不平衡数据的主动学习算法Balance adjustment Active Learning(简称Ba-AL).每次迭代结束检查训练集样本平衡度,对不平衡训练集进行聚类并剔除冗余样本,保持训练集的平衡,从而提高分类效果.UCI数据集及真实的遥感影像数据集仿真结果表明,该方法可以获得较好的分类效果,达到目标正确率所需的最少训练样本数更少,算法效率更高,数据利用指标更优越.

An active learning algorithm called Balance adjustment Active Learning(Ba-AL for short)is proposed for imbalanced data.At the end of each iteration,the sample balance of the training set is checked,and the unbalanced training set is clustered and redundant samples are eliminated,so as to maintain the balance of the training set and improve the classification effect.Simulations on UCI datasets and real remote sensing image datasets show that this method can achieve better classification results,the minimum number of training samples required to achieve the target accuracy rate is less,and the algorithm is more efficient and data utilization indicators are better.