Year: 2025
Type: article
Abstract: Article is a study of methods for evaluating the effectiveness of feature selection algorithms and proposes a new methodology for their evaluation. It is noted that existing methods and approaches to assessment do not always adequately reflect the actual efficiency of algorithms, especially when applied to real problems. The article comprehensively discusses various opinions of researchers on the issues of internal and external validity of existing evaluation methods, assesses the impact of different parameters, including data volume, differences in algorithm implementations, and other factors. The authors propose a new integrated approach to evaluating the effectiveness of algorithms, which includesa set of indicators such as resource costs, stability, and task solution quality. One of the peculiarities of the proposed approach is the use of artificially generated data, which allows considering specific characteristics of real data and evaluating algorithms under controlled conditions. This makes it possible to more accurately determine their efficiency and reliability. The article also contains the results of preliminary testing of the proposed methodology using artifi cial data. The analysis of these results demonstrates the advantages of the new approach over traditional evaluation methods. In particular, it was found that the new methodology allows more accurately evaluating the stability and quality of algorithm performance, which is crucial for making decisions about choosing an appropriate algorithm for specific tasks. In conclusion, the authors emphasize the need for further development and expansion of the proposed methodology, as well as its adaptation for solving various types of tasks. They also point out the prospects for integrating this methodology into specialized software packages, which will make it accessible to a wide range of users and accelerate the implementation of innovative algorithms in practice.
Authors: A. D. Cheremuhin, A. D. Rein
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4414148564