Premium
Mining abnormal patterns from heterogeneous time‐series with irrelevant features for fault event detection
Author(s) -
Fujimaki Ryohei,
Nakata Takayuki,
Tsukahara Hidenori,
Sato Akinori,
Yamanishi Kenji
Publication year - 2009
Publication title -
statistical analysis and data mining: the asa data science journal
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.381
H-Index - 33
eISSN - 1932-1872
pISSN - 1932-1864
DOI - 10.1002/sam.10030
Subject(s) - anomaly detection , computer science , anomaly (physics) , data mining , fault (geology) , event (particle physics) , series (stratigraphy) , multivariate statistics , sequence (biology) , pattern recognition (psychology) , time series , artificial intelligence , algorithm , machine learning , physics , genetics , seismology , biology , condensed matter physics , quantum mechanics , geology , paleontology
We address the issue of detecting fault events in multivariate time series. We suppose the following realistic and challenging situations: (A) the features to which multivariate time series correspond are heterogeneous; (B) only a small number of examples of fault events are available in advance relative to a large number of normal examples; and (C) many features irrelevant to fault events are included. To resolve these situations, detecting faults specifically in machine systems such as automobile, train, etc. is usually required. We propose an algorithm to resolve the issue. Key ideas in it include: (1) transforming the time‐series for each feature into a sequence of anomaly scores, in order to map heterogeneous features to homogeneous ones (an anomaly score indicates the degree of anomalousness relative to an ordinal sequence) and then representing the pattern of a fault event in terms of fault anomaly score vectors; (2) selecting features specifying a fault event by means of iterative optimization using both normal and fault anomaly score vectors. We then monitor the degree of anomalousness with respect to target anomaly score vectors by comparing them with the abnormal patterns. We demonstrate the effectiveness of our proposed algorithm through an application to an actual automobile fault diagnosis data set as well as an artificial dataset. © 2009 Wiley Periodicals, Inc. Statistical Analysis and Data Mining 2: 1‐17, 2009