Кібернетика та комп'ютерні технології (Dec 2023)

Algorithm for Classification of Patient Groups Based on the LDA Method

  • Vitalii Budnyk,
  • Mykola Budnyk

DOI
https://doi.org/10.34229/2707-451X.23.4.9
Journal volume & issue
no. 4
pp. 76 – 83

Abstract

Read online

Introduction. The article is devoted to the development of the algorithm of classification of several groups of patients using the LDA method at certain stages of its work. The algorithm allows achieve high discrimination value, to work with a variety of parameters and to separate more than two groups of patients. It is used for the analysis of biomedical data, namely for the array of indicators of various nature described healthy persons and persons with the detected disease. The purpose of paper is to develop an algorithm to increase the power of discrimination of an array of medical data that may include hundreds of healthy people and patients with connective tissue diseases, and each person is described by hundreds of parameters. Results. The authors proposed an algorithm for increasing classification accuracy based on parameter selection using sequential multivariate LDA tests. An example of the application of the algorithm to the task of analyzing a fairly large array of data (501 individuals: 295 healthy, 206 patients, 240 parameters) and identifying informative parameters for diagnosing children with connective tissue diseases is given. As a result of the application of this algorithm, a discriminant function and a decision rule were obtained, which allows achieve an average accuracy of discrimination over the entire set of parameters of 85 %. In addition, the algorithm is applied for two separate groups of parameters - blood indicators and biochemical analysis, while the average accuracy of discrimination reaches 84 % and 90 %, respectively. Conclusions. The algorithm of classification groups of patients with the use of LDA has been developed, which allows achieve high accuracy of discrimination. The results of its application in solving the real data set are given. The results of its application to solving the task of classification of medical data show its ability to improve classification accuracy.

Keywords