Data Mining In Non-Stationary Multidimensional Time Series Using A Rule Similarity Measure

Time series analysis is a wide area of knowledge that studies processes in their evolution. The classical research in the area tends to find global laws underlying the behaviour of time series, the contemporary data mining in time series mainly focuses on the mining of local rules. In plenty of the real world applications it is extremely important for a time series data mining algorithm to perform correctly on the non-stationary time series. Moreover, in plenty of the real world applications it should have a "continuous" change of the data mining algorithm's parameters. In this paper a novel approach for the mining of slowly changing rules is introduced. This approach uses a model of rule that allows mining both universal and local rules. Moreover, the approach is able to work with non-stationary time series and performs the mining of the slowly changing rules. The key structure for measuring the "speed of change" is the rule similarity measure. It is used as ...

• Type: Conference
• Year: 2008
• Where: IADIS
• Authors: Nikolay V. Filipenkov
• Keywords: Data Mining, Data Mining Algorithm, IADIS 2008, Internet Technology, Time Series