ANALYSIS OF RELATIONSHIPS BETWEEN TRAINING SAMPLE OBJECTS ON MANIFOLDS OF LATENT FEATURE SETS
DOI:
https://doi.org/10.65164/43sjv566Keywords:
latent features, dimensionality reduction, hierarchical agglomerative clustering, pattern recognition, compactness measure.Abstract
This paper proposes an approach to pattern recognition based on latent feature sets
formed through hierarchical agglomerative clustering using additive and multiplicative principles.
The compactness of the resulting feature sets is evaluated via a connectivity ratio based on a
hypersphere system. A computational experiment on genomic DNA data demonstrates that the
multiplicative principle outperforms the additive one, achieving a recognition accuracy of 96.15%
with only 5 latent features, confirming its effectiveness for solving dimensionality reduction problems
in data mining tasks.
References
[1]. Ignatyev N. A. Structure Choice for Relations between Objects in Metric Classification
Algorithms // Pattern Recognition and Image Analysis. 2018. V. 28. № 4. P. 590–597.
[2]. Ignatev N. A. and Rahimova M. A. Formation and Analysis of Sets of Informative Features of
Objects by Pairs of Classes // Scientific and Technical Information Processing, 2022, Vol. 49,
No. 6, pp. 439–445.
[3]. Saidov D.Y. Data visualization and its proof by compactness criterion of objects of classes //
(IJISA), 2017. Vol 9. № 8. P. 51–58.
[4]. Sudmant P. H. et al. "An integrated map of structural variation in 2,504 human genomes,"
Nature vol. 526, pp. 75–81, Oct. 2015. doi:10.1038/nature15394.
