MAPS: A Method for Identifying and Predicting 
Aberrant Behavior in Time Series
=============================================

by  Evangelos Kotsakis and Antoni Wolski

Proc. 14th Internat. Conf. on Industrial and Engineering 
Applications of Artificial Intelligence and Expert Systems 
(IEA/AIE-2001), June 4-7, 2001, Budapest, Hungary.

Abstract

We present a method for inducing a set of rules from time 
series data, which is originated from a monitored process. 
The proposed method is called MAPS (Mining Aberrant Patterns 
in Sequences) and it may be used in decision support or 
in control to identify faulty system states. It consists 
of four parts: training, identification, event mining and 
prediction. In order to improve the flexibility of the event 
identification, we employ fuzzy sets and propose a method 
that extracts membership functions from statistical measures 
of the time series. The proposed approach integrates fuzzy 
logic and event mining in a seamless way. Some of the existing 
event mining algorithms have been modified to accommodate the 
need of discovering fuzzy event patterns.