Local neighborhood rough set

• Published in: Knowledge-based systems 2018-08, Vol.153, p.53-64
• Main Authors: Wang, Qi, Qian, Yuhua, Liang, Xinyan, Guo, Qian, Liang, Jiye
• Format: Article
• Language: English
• Subjects: Algorithms; Attribute reduction; Big Data; Cloud computing; Cognition & reasoning; Concept approximation; Data management; Data mining; Limited labeled data; Local neighborhood rough set; Reduction; Rough set; Rough set models; Set theory
• ISSN: 0950-7051; 1872-7409
• DOI: 10.1016/j.knosys.2018.04.023

•Local neighborhood rough set (LNRS) is proposed.•The LNRS model can handle big data with numeric attributes and limited labels.•The corresponding concept approximation and attribute reduction algorithms have linear time complexity.•The study provides a bridge between neighborhood rough set and local rough set. With the advent of the age of big data, a typical big data set called limited labeled big data appears. It includes a small amount of labeled data and a large amount of unlabeled data. Some existing neighborhood-based rough set algorithms work well in analyzing the rough data with numerical features. But, they face three challenges: limited labeled property of big data, computational inefficiency and over-fitting in attribute reduction when dealing with limited labeled data. In order to address the three issues, a combination of neighborhood rough set and local rough set called local neighborhood rough set (LNRS) is proposed in this paper. The corresponding concept approximation and attribute reduction algorithms designed with linear time complexity can efficiently and effectively deal with limited labeled big data. The experimental results show that the proposed local neighborhood rough set and corresponding algorithms significantly outperform its original counterpart in classical neighborhood rough set. These results will enrich the local rough set theory and enlarge its application scopes.