Practical evaluation, issues and enhancements of applied data mining

Abstract

The wide popularity of Data Mining stimulated further expansion of spectrum of its methods and tools. However, the studies relating to this area usually focus on its scientific perception, and neglect other factors, being important in the practical evaluation of Data Mining and its applications, including stability of results, easiness of algorithms usage, demands on computational resources or speed factors. In this paper, we will provide the most significant limitations of applied Data Mining. Later, we will investigate how to perform the full evaluation of Data Mining methods and their results - beyond commonly used limited evaluation of results quality. The analysis will be done for two most popular types of tasks performed in Data Mining projects, i.e. feature selection and classification, but its results might be applied to other tasks. We will also provide information crucial to software applications based on Data Mining, including flexibility of knowledge representation forms. © 2011 Springer-Verlag.