Biological Data Mining and Its Applications in Pulmonology

Abstract

The processing of diagnostic data in pulmonology is complicated for a doctor due to the need to analyze many indicators, the relationships between which can be complex, and the degree of influence on the diagnostic result can be different. Traditionally, general clinical, biochemical, and questionnaire methods are used to support making a diagnosis. They allow describing the state of the bronchopulmonary system by a variety of indicators. The modern practice of laser correlation spectroscopy makes it possible to expand various indicators. Still, their values are represented by sets of one-dimensional distribution diagrams and are not convenient for analysis. Therefore, we investigated the feasibility of classifying the values of 32 biophysical indicators obtained by laser correlation spectroscopy in this work. We first performed data visualization and found that the classes of diagnosed diseases did not have a clear separation but were separated from the normal state. We then examined the results of classifying the data using three algorithms – naive Bayes, logistic regression, and random forest. We conclude that the most appropriate algorithm is logistic regression. The work value lies in expanding the set of diagnostic indicators due to the high-precision results of the classification of biophysical indicators, which increases the objectivity of the diagnosis of pulmonological diseases.