BUILDING AN ACCEPTANCE SUPPORT SYSTEM MEDICAL DECISIONS BASED ON HETEROGENEOUS ENSEMBLE CLASSIFIER

Based on the analysis of diagnostic methods used to make a diagnosis by a doctor in accordance with the requirements of evidence-based medicine, as well as methods of constructing diagnostic decision rules in medical decision-making support systems, two approaches to the formulation of a diagnostic conclusion are distinguished. The first is the traditional approach of medical practitioners based on the identification of disease symptoms and analysis of the structure of symptom complexes, taking into account the experience of clinicians in this subject field of medicine, which is covered in medical handbooks, and it can be considered an expert assessment of each disease. The second is a formalized approach based on the synthesis of a probabilistic classifier that is trained on a training sample. Since these approaches consider the same problem from different points of view, their joint use is promising. Implementation of different types of classifier models and aggregation of their results is possible when implementing a heterogeneous ensemble classifier. The goal of the study is the synthesis of a heterogeneous ensemble classifier that takes into accounts both the expert and probabilistic components of the diagnosis process. In the paper, a mathematical model of a heterogeneous ensemble classifier is developed and the choice of its components is substantiated. As a probabilistic component, widely used classification methods are selected: the method of comparison with a prototype, the method of K-nearest neighbors, and the method of potential functions. Formalized expert information on the structure of symptom complexes by presenting the symptom complexes of diseases with numerical intervals of values "below normal", "normal", "above normal" of linguistic variables. Considered options for taking into account expert assessments regarding the structure of symptom complexes in the ensemble classifier. The decision support system was designed and a comprehensive check of the developed system was carried out on real medical data, which confirmed the effectiveness of the system.