Intelligent Data Mining

Sunday, 9:00 - Noon

(Pavilion Suite 1)

 
Dr. Iveta Mrazova, Charles University, Czech Republic
 

Overview: This tutorial provides an overview of principal concepts and techniques applicable to data mining. The subject of data mining consists mainly in exploring and analyzing large quantities of data with the aim to discover their mutual relationships. In this context, intelligent refers to systems that can interact with their environment and that can adapt themselves to changes both in the feature space and in time. Emphasis will be placed on adaptive methods developed for data mining and their capability to detect meaningful novel patterns. The ability to detect significant input patterns and to identify their characteristic features can be used both for training and optimizing the structure of the model at hand and for improving its performance - generalization abilities, robustness, storage capacity, etc. Simultaneously, it might help to gain insight into the structure of the processed data and to detect irregularities and errors in it.

To the tasks well-suited for data mining belongs classification, estimation, prediction, affinity grouping, clustering and description. There are known several types of data mining techniques – among others cluster detection, memory-based reasoning, market basket analysis, decision trees, link analysis, artificial neural networks, fuzzy logic based systems, genetic algorithms. Many of these techniques were introduced few years or decades ago, mostly in the area of computer science and artificial intelligence. Data mining algorithms typically require multiple passes over large volumes of data and many of them are computationally intensive. Anyway, several trends are increasing the necessity of powerful data mining tools - an increasingly service-based economy, the advent of mass customization and a competitive advantage of appropriate information.

Applications of data mining techniques reach a wide variety of fields - economics, artificial intelligence (AI), databases, Web technologies, medicine and statistics. The choice of a particular combination of techniques to apply in a given situation depends on both the nature of the data mining task to be accomplished and the nature of the available data. Appropriate visualization of mutual relationships among the data enables qualified decision making and reasoning. Unfortunately, for some models it is relatively complicated to explain and visualize what they are doing. For other purposes, it might be useful to extract a clear set of simple rules providing insight into how is a particular model working. A similar requirement represents an easy reusability of the applied model.

Instructor's Background: Dr. Iveta Mrázová teaches courses on Data Mining and Artificial Neural Networks at Charles University in Prague, Czech Republic. She received the M.S. degree from the Friedrich-Schiller-University, Jena (Germany) in 1989 and the Ph.D. degree from the Institute of Computer Science of the Czech Academy of Sciences in 1997. She published numerous research papers in the area of artificial neural networks, pattern recognition and image processing. In 1996, she received the Annual Prize of the Bolzano Foundation for a collection of original publications “On the Internal Knowledge Representation in Neural Networks.” In 2000, her paper "Generalized Relief Error Networks" won the first runner up award in “Theoretical developments in computational intelligence” of ANNIE´2000 (St. Louis, USA). In 2001, the Union of Czech Mathematicians and Physicists and the Czech Society for Mechanics awarded her for “outstanding work in the field of computer science” by the Prize of Prof. Babuška. During September, 2002 – June, 2003, she joined the Engineering Management Department, University of Missouri Rolla as a Fulbright Visiting Scholar.