The Issue of Efficient Generation of Generalized Features in Algorithmic Classification Tree Methods
The paper studies the basic issue of methods of constructing models of algorithmic classification trees which is the problem of generating generalized features in their structure. There has been suggested a simple and efficient method of approximating the
- PDF / 320,631 Bytes
- 16 Pages / 439.37 x 666.142 pts Page_size
- 69 Downloads / 190 Views
stract. The paper studies the basic issue of methods of constructing models of algorithmic classification trees which is the problem of generating generalized features in their structure. There has been suggested a simple and efficient method of approximating the initial training dataset with the help of a set of generalized features of the type of geometric objects in feature space of the current application problem. This approach allows ensuring the necessary accuracy of classification trees, reducing the structural (constructional) complexity and achieving the appropriate performance indicators of the model. Based on the proposed algorithmic scheme of the training set coverage there has been developed the corresponding software which enables the work with a set of different-type applied problems of data processing. Hence, a simple, efficient and economical approximation of the initial training set provides the appropriate speed, level of complexity of the classification scheme, which ensures the simple and complete recognition (coverage) of sets of discrete objects. Keywords: Discrete objects recognition · Approximation by means of geometric objects · Recognition function · Generalized feature · Classification trees
1
Introduction
The research offers a solution to the problem of pattern recognition in the deterministic variant based on the use of the methodology (concept) of the logical (algorithmic) tree method [2,6,7,15,18–23,31,32]. The specificity of methods of classification trees (decision trees) enables creating on its basis software systems for automating the process of constructing new algorithms and recognition systems in general. And within the approach of an algorithmic classification tree (ACT) [18], whose main idea is to approximate the initial training dataset with the help of a set of independent classification algorithms, there arises the central question regarding the issue of generating generalized features (the vertices of the classification tree structure) [20]. In this way an opportunity to effectively c Springer Nature Switzerland AG 2020 S. Babichev et al. (Eds.): DSMP 2020, CCIS 1158, pp. 98–113, 2020. https://doi.org/10.1007/978-3-030-61656-4_6
Generalized Features in Algorithmic Classification Tree Methods
99
use the long-term experience, accumulated in the theory and practice of pattern recognition, giving already known algorithms and recognition programs a second life, is fulfilled. In this respect, the tree method completes the branch feature selection methodology [12]. Therefore, a simple and efficient method of approximating a training set (or part of it) by means of geometric objects is proposed. The idea of the developed algorithm for generating generalized features (GFs) [20] still lies in the approximation of a certain class by means of the sequence of certain generalized features [30] (hyperparallelepipeds). With regard to its algorithmic implementation, it should be noted that it develops the work [29] that was not devoid of certain systemic constraints (fixed orientation in space, the im
Data Loading...
