Open AccessPBWA: A Provenance‐Based What‐If Analysis Approach for Data Mining Processes
Author(s) -
Ke Jie,
Dong Hongbin,
Tan Chengyu,
Liang Yiwen
Publication year - 2017
Publication title -
chinese journal of electronics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.267
H-Index - 25
eISSN - 2075-5597
pISSN - 1022-4653
DOI - 10.1049/cje.2017.06.003
Subject(s) - metadata , computer science , provenance , data mining , data science , information retrieval , world wide web , geology , petrology
This paper presents a Provenance‐based what‐if analysis approach (PBWA) for data mining processes, so decision makers can examine the latest mining result under hypothetical business contexts. It fills the gap that data mining only reveals past status of enterprises with historical data. Provenance information is a kind of metadata of data mining processes. PBWA uses it to identify relevant operation path and intermediate results that is affected by hypothetical business contexts. It refreshes the mining result by partially rerunning the affected portions. Different from previous studies only for relational operations, PBWA can take more general operations into account. Besides, it focuses on the whole mining processes. Experiments demonstrate that when the affected ratio is less than 74% and 87% in different contexts, PBWA can achieve better time performance.