Meta-knowledge Management in Multistrategy Process-oriented Knowledge Discovery Systems

Current electronic data repositories are growing quickly and contain big amount of data from commercial, scientific, and other domain areas. The capabilities for collecting and storing all kinds of data exceed the abilities to analyze, summarize, and extract knowledge from this data. Knowledge discovery systems (KDSs) use achievements from many technical areas, including databases, Data Mining (DM), statistics, AI, machine learning, pattern recognition, high performance computing, management information systems (MIS), decision support systems, and knowledge-based systems. Knowledge discovery is an innovative approach to information management and is associated commonly with the nontrivial process of identifying valid, novel, potentially useful, and ultimately understandable patterns and relations in large databases (Fayyad, 1996). Numerous data mining techniques have recently been developed to extract knowledge from large databases. Since present-day KDSs are armed with a number of available techniques to process data; and, potentially, there are many possible combinations of these techniques to construct a DM strategy for mining a current problem. In a real problem-solving situation it is not computationally feasible to apply every DM strategy. Therefore, dynamic selection of data mining methods in knowledge discovery systems has been under active study (see, for example, (Tsymbal, 2002)). However, at least two contexts of dynamic selection can be distinguished. First, the so-called multi-classifier systems that apply different ensemble techniques (Dietterich, 1997). Their general idea is usually to select one classifier on the dynamic basis taking into account the local performance (e.g. generalisation accuracy) in the instance space. Second, multistrategy learning that applies a strategy selection approach which takes into account the classification problemrelated characteristics (meta-data). We are interested in the second context in this study. Selection of the most appropriate DM technique or a group of the most appropriate techniques is usually not straightforward. Many empirical studies are aimed