A data-scattering-preserving adaptive self-organizing map. (October 2021)

• Record Type: Journal Article
• Authors: Olszewski, Dominik
• Is Part Of: Engineering applications of artificial intelligence. Volume 105(2021)
• Publication Date: 2021-10
• Subjects: Self-organizing map -- Adaptive self-organizing map -- Data visualization -- Data clustering -- Inner-cluster variance -- Neighborhood width; Engineering -- Data processing -- Periodicals; Artificial intelligence -- Periodicals; Expert systems (Computer science) -- Periodicals; Ingénierie -- Informatique -- Périodiques; Intelligence artificielle -- Périodiques; Systèmes experts (Informatique) -- Périodiques; Artificial intelligence; Engineering -- Data processing; Expert systems (Computer science); Periodicals; 620.00285
• DOI: 10.1016/j.engappai.2021.104420
• Languages: English
• ISSNs: 0952-1976

Abstract

Abstract: We propose a novel improved adaptive version of the Self-Organizing Map (SOM). Our enhancement of SOM adapts to the input data scattering, in this way, preserving the principal input data structure. In order to capture and extract the data scattering from the input dataset, we propose carrying out a preliminary input data clustering. As a result of forming the clusters in the input data space, the inner-cluster variances are calculated. The inner-cluster variances are subsequently employed as the basis for the determination of the neighborhood widths of the SOM's Best Matching Units (BMUs). The method, introduced in this paper, has been a subject to the empirical evaluation carried out using the three real-world datasets of different size, dimensionality, and data type, representing three various experimental fields. Our proposal has been assessed on the basis of a comparison with seven reference data visualization approaches. The introduced method appeared to be superior over the remaining investigated data visualization techniques, and consequently, its effectiveness, usefulness, and accuracy have been verified and confirmed. Highlights: A Self-Organizing Map (SOM) adapting to the input data scattering is proposed. The input data scattering is expressed numerically by the inner-cluster variances. The inner-cluster variances are obtained after the initial input data clustering. The inner-cluster variances controle the neighborhood widths of the SOM's neurons. The experiments on real-world data confirmed the effectiveness of our approach.