Naslov
REFII model-Model for recognition patterns in time series

Autori
Klepac, Goran

Vrsta, podvrsta i kategorija rada
Sažeci sa skupova, sažetak, znanstveni

Izvornik
Program and astracts, 20th International conference METHODOLOGY AND STATISTIC / - Ljubljana, 2002, 53-55

Skup
20th International conference METHODOLOGY AND STATISTIC, University of Ljubljana

Mjesto i datum
Ljubljana, Slovenija, 15.09.2002. - 18.09.2002

Vrsta sudjelovanja
Predavanje

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
Data mining; time series

Sažetak
REFII model is an authorial mathematical model for recognition patterns in time series. It is important to say that REFII model is not a closed system, meaning that we have a finite set of methods. It is in the first place a model for a transformation of values of time series, which prepares data that are used by different sets of methods in a domain of problem space in order to solve problems. The purpose of the model is to: discover seasonal oscillation, discover cyclic oscillation, discover rules from time series, discover episodes from time series, discover similarity of time segments, discover correlation between time segments, discover rules from in domain of finances from time series, connect time series and standard data mining methods, analyze time series with the help of data mining methods (clustering of time segments, classification of time segments) The mathematical background is focused on two basic elements of time series: shape and area beneath the curve. All submodels, which solve specific problems in different domains, use these two elements in all algorithmic procedures. REFII model is able to analyze every time series, which is represented by values. In the first step of the analysis, after preparing the original data, we make a transformation of time series in REFII model values. The next step is to select an appropriate method inside REFII model to analyze data. The selection of the method is determined within the scope of the analysis. Methods could be focused on discovering: seasonal oscillation, cyclic oscillation, hidden rules, episodes, similarity of time segments, correlation of time segments, clusters, or links between time segments. The advantage of REFII model is its possible application in many different areas like finance, medicine, voice recognition, face recognition, text mining. As an illustrative example of how the model could be efficient, the accent in the presentation will be put on its application in banking and finance. REFII model could be successful in discovering rules of bank clients' behavior, when their behaviors depend on a time dimension. These rules could be connected with behavior when using transactional services, behavior with money management, prediction of using a new banking service in an expected period of time, or similarity between a client and a market group. We could integrate REFII model in a query language for time series, and in that case we would get a powerful tool for creating complex algorithmic procedures based on pattern recognition methodology. A large number of different methods could be used as elements of the query language, but we could also use every of them like a single method for a single analysis. The results of the analysis could be presented in IF-THEN form, which means that it is possible to use an expert system shell and metarules to analyze knowledge from time series analysis. REFII model is an open system, which could be used like a connection between time series and other mathematical, statistical, and data mining methods. Main reasons why the system has been developed are: there is no unique methodology in data mining for time series analysis, the existing system for time series analysis, that is a part of statistics, doesn't provide the answers in the spirit of the data mining methodology, and doesn't give the solutions on how to connect the standard data mining methods with time series, when we have a problem in a domain of specific business area and time series, we have to develop a whole new system whose concept is often not compatible with other similar solutions, to find an efficient system that will be able to predict events in time series, search for similar patterns, search for seasonal and cyclic oscillation, find rules from time series, find correlation between time segments, or two or more time series, to use the potential of time series in medicine (EEG - searching for patterns of mental illness, EKG - searching for patterns of heart illness), face recognition (REFII model in 3D space), voice recognition, text mining, and multimedia data mining. The first results of using REFII model are good. The concept could exist as an application that integrates a variety of sub-models, as a query language for time series, or as modules that are integrated within other applications.

Izvorni jezik
Engleski

Znanstvena područja
Informacijske i komunikacijske znanosti

POVEZANOST RADA