Incremental multiple instance outlier detection

Autor:	Shifeng Weng, Zhigang Wang, Zeng-Shun Zhao, Changshui Zhang
Rok vydání:	2014
Předmět:	Data stream Local outlier factor business.industry Computer science Pattern recognition computer.software_genre Task (computing) Artificial Intelligence Outlier Anomaly detection Artificial intelligence Data mining business computer Software
Zdroj:	Neural Computing and Applications. 26:957-968
ISSN:	1433-3058 0941-0643
Popis:	I-MLOF algorithm is an extension of local outlier factor (LOF) algorithm in multiple instance (MI) setting. The task of I-MLOF is to identify MI outlier. However, I-MLOF algorithm works in batch mode, where all samples must be provided for once. In some real applications such as industrial detection and traffic monitoring, MI outlier is required to be identified from data stream. The batch-mode outlier detection methods usually cannot be applied directly to these applications. In this paper, an incremental MI outlier detection algorithm "Inc I-MLOF" is proposed. MI outlier detection can be done for sequentially arrived data with Inc I-MLOF. We prove theoretically that Inc I-MLOF achieves the equal result to that of I-MLOF. The experimental results illustrate Inc I-MLOF achieves good performance on several synthetic and real data sets.
Databáze:	OpenAIRE
Externí odkaz:	https://explore.openaire.eu/search/publication?articleId=doi_________::3d1926857a3f2d54c2b4818befe57216 https://doi.org/10.1007/s00521-014-1750-6 Zobrazit plný text záznamu Full text from SpringerLink