Dynamic Data Mining: Synergy of Bio-Inspired Clustering Methods

Abstract

Dynamic data mining (DDM) comprises advantages of static methods used to reveal implicit structure of classes and at the same time benefits from high quality results obtained in the field of time series analysis. Clustering problem is recognized to be the most crucial in almost any knowledge domain: telecommunications and networking, nanotechnology, physics, chemistry, biology, health care, sociology, economics, etc (Aliev & et. al., 2008; Ceylan & et. al., 2009; Chee & Schatz, 2007; Ghosh & et. al.; 2008; Pedrycz & Weber, 2008; Xu & et. al., 2010). Scientists are in chase of new materials and new decision making techniques that manage data, information, devices or people on the fly. Centralized management techniques are mainly ineffective if we need to operate heaps of redundant information under uncertainty in real time. Decentralized control of interconnected elements, called networks, collectives, colonies, ensembles, maps is based on self-organization and bioinspired principals that underlie amazing effects applied in highly interdisciplinary environment. In the paper we extend the thorough comparative analysis of bio-inspired methods provided in resent research (Blum & Merkle, 2009; Budyan & et. al., 2009; Dressler & Akan, 2010) for benefit of clustering problem. Under consideration are the following bio-inspired approaches, used to reveal implicit data structures: small-world networks, ant-based networks, fuzzy logic, neural networks, chaotic map lattices, classical data mining, selforganizing maps. The general view on advantages and delimitations on various bio-inspired methods combinations is proposed in the form of a decision tree. There are so many combinations of bio-inspired methods (Crespo & Weber, 2005; Georgieva & Klawonn, 2008; Jaimes & Torra, 2010; Kaiser & et. al., 2003, 2007; Li & Shen, 2010, Sussillo & Abbott, 2009) with various extent of effectiveness that we tried to propose a systematic approach to reveal best practices. We state that significant advantages in terms of high quality clustering results can be obtained when complexity of both structure and dynamics is commensurable with complexity of problem. This is possible when we tune the harmony of more than two or three techniques. Distributed manner of decision-making processes in nature dictate multiform compensation of possible ineffective functioning of separate system’s element by collective dynamics of all other elements. Detailed analysis of simultaneous clustering techniques application within one method is given on the example of chaotic neural network (Benderskaya & Zhukova 2008, 2009). The

3 Figures and Tables

Cite this paper

@inproceedings{Benderskaya2012DynamicDM, title={Dynamic Data Mining: Synergy of Bio-Inspired Clustering Methods}, author={Elena N. Benderskaya and Sofya V. Zhukova}, year={2012} }