A Selection Metric for semi-supervised learning based on neighborhood construction
作者：
Highlights：
• We show that the confidence-based selection metric may not exploit a set of informative unlabeled data to improve the classification performance.
• The new selection metric is proposed to semi-supervised self-training considering close data points to decision boundary.
• The proposed selection metric uses the agreement between classifier prediction and the proposed neighborhood construction.
• We propose a new sample selection from the unlabeled data using a neighborhood construction algorithm.
• The proposed neighborhood construction algorithm employs Apollonius circle and dense data points.
摘要
•We show that the confidence-based selection metric may not exploit a set of informative unlabeled data to improve the classification performance.•The new selection metric is proposed to semi-supervised self-training considering close data points to decision boundary.•The proposed selection metric uses the agreement between classifier prediction and the proposed neighborhood construction.•We propose a new sample selection from the unlabeled data using a neighborhood construction algorithm.•The proposed neighborhood construction algorithm employs Apollonius circle and dense data points.
论文关键词：Apollonius circle,Semi-supervised classification,Self-training,Support vector machine,Neighborhood construction
论文评审过程：Received 23 August 2020, Revised 2 November 2020, Accepted 24 November 2020, Available online 22 December 2020, Version of Record 22 December 2020.
论文官网地址：https://doi.org/10.1016/j.ipm.2020.102444
