Knowledge discovery in databases with adaptive control systems. The development of a data mining methods tool-bar basing on neuro-fuzzy systems (Q2703375)

Knowledge Discovery in Databases (KDD) is becoming a more and more important part of business intelligence. This volume is concerned with the use of adaptive rule-based systems for pattern classification and tries to cover the topic completely. Classical statistical methods are often insufficient because of huge data volumes, special requirements for their application, bad scaling, missing robustness and difficulties to interpret the results. The author proposes neuro-fuzzy systems as a basis for a data mining and classification system. His concept comprises five steps: data selection and transformation, data preprocessing, rule extraction by a neuro-fuzzy system, and pattern postprocessing. This concept is the topic of chapter 3. Chapter 1 includes preparatory material, while chapter 2 first gives short introductions into several classical methods like discriminatory analysis, nonparametric estimation, descriptive methods as well as methods from artificial intelligence like neural networks, decision trees and rule induction. NEWLINENEWLINENEWLINEThe main part of chapter 2 however is concerned with the classification of adaptive rule based systems. Fuzzy logic and neural networks are first studied individually before they are combined. Several implementations of various methods are described here, including the NEFCLASS prototype, developed at the university of Magdeburg, which is subsequently explored under various aspects. NEWLINENEWLINENEWLINEIn chapter 3 the author stresses the importance of pre- and postprocessing data resp. rules. The work ist the author's thesis and accordingly written. This is apparent here (and in chapter 2) because the author always carefully compares and judges different methods based on foreign and own empirical investigations. Special attention is payed to the problem of missing values. Where appropriate pseudo-code for algorithms is given, time complexities are discussed. At last a complete system for KDD using various software systems (some available as freeware) is proposed. NEWLINENEWLINENEWLINEFinally chapter 4 contains three case studies of data mining using the proposed system. There are many illustrations, diagrams and formulas without going into complete mathematical detail. The bibliography is vast. The reader will find enough theoretical background to judge the methods and is also equipped with a guide for solving problems in practice.