The area of data mining and knowledge discovery is inherently associated with databases. Data mining methods are used in the process of knowledge discovery to reveal new pieces of knowledge from large databases. One of the stages in that process is a feature selection. A feature selection is usually meant as a process of finding a subset of features from the original set of features forming patterns in a given data set, optimal according to the defined goal and criterion of feature selection. The aim of this paper is to present the main functionalities of the Data Mining Exploration System (DMES) [4] and to explain its usefulness to the feature selection tasks. The DMES is an integrated software system that incorporates many algorithms which can be used in data mining. It is currently being developed in the University of Rzeszow. We describe in short its recent version (1.2) and the algorithms that had already been implemented. The DMES allows to visualize, split, preprocess, analyze, classify and reduce decision tables. To show one of the possibilities provided by the DMES system, we used it for the feature selection [3], [7], [8], [10], [12] task. Feature selection is performed by the most recent addition to the DMES system which are RBFS1, RBFS2 and ARS algorithms [14]. These feature selection methods are designed mainly for the multiple classifiers systems with homogeneous classifiers [2], [5], [6]. Homogeneous classifiers require many different subsets of the data set. The problem of finding the best subsets of a given feature set is exponentially complex. We use RBFS (Reduct Based Feature Selection) algorithm [15] and its two modifications called RBFS1 and RBFS2 to select optimal subsets of the feature set for multiple classifiers. RBFS algorithms are quite complex computationally because they use all decision-relative reducts [11] of a given decision table.
展开▼
