TITLE:
Reducing Data Chaos and Partitioning the Training Sample into Macro-Features in Classification Problem
AUTHORS:
Vladimir Shats
KEYWORDS:
Reducing Data Chaos, Hidden Variables, Functional Relationship “Feature-Class”
JOURNAL NAME:
Journal of Intelligent Learning Systems and Applications,
Vol.18 No.2,
May
26,
2026
ABSTRACT: This paper is devoted to revealing some features of machine learning problems and developing a new approach to solving them. It is based on the application of information processing technology by animal sensory systems, each of which perceives information of only a certain type. Therefore, computational operations for solving the classification problem are performed mainly for individual features of objects, although they are usually carried out for objects as a whole. This approach ensures the simplicity of the algorithm and the ability to order the features by sorting in non-decreasing order their values, which leads to a decrease in the entropy and chaos of the data. It has been established that ordered features are hidden variables that allow us to detect the functional relationship “feature-class” and to partition any training sample into macro-features, which are ordered features of objects of a certain class. Classification of any object in test sample comes down to calculating the frequency of occurrence of its feature values in the nearest neighborhood of ordered feature values of the corresponding macro-feature of a certain class. The object class corresponds to the maximum of the average value of this frequency. Applying of ordered features opens up the prospect of new types of neural networks. The article also discusses the application of an ordered data matrix to solve problems of partitioning a set into clusters of objects with common properties.