USING THE KOHONEN NEURAL NETWORK TO CLASSIFY DISEASES OF THE CARDIOVASCULAR SYSTEM

This article presents the results of research on the use of the Kohonen neural network for the purpose of classification of the cardiovascular system diseases. A one-dimensional grid architecture was used with dynamically variable number of network classes at its creation stage. The network was trained on a dataset containing 1,200 records, each of which was correlated with one of the four diagnoses: normal heart rate, heart failure, atrial fibrillation, arrhythmia. The results of the ECG procedure were considered as records. The dependence of the accuracy of the classification of input data by the network on the network parameters was assessed, and the values of the average classification accuracy were obtained. Optimal parameters for training a neural network using self-learning and vector quantization algorithms were provided, and the influence of the association of network classes with multiple diagnoses on learning efficiency and the accuracy of record classification were studied.