Applying the Kohonen Neural Network to Solve the Problem of Determining Diseases of the Cardiovascular System Based on the Results of Electrocardiography

The paper researches the use of the Kohonen neural network to solve the problem of determining cardiovascular diseases based on the results of electrocardiography. Self-study and teacher training were applied, the sample included records corresponding to the following diagnoses: normal sinus rhythm (NSR), arrhythmia (ARR), atrial fibrillation (AFF) and heart failure (CHF). The best results in 96% accuracy were shown by a neural network created with a threshold error value of 60, a starting learning coefficient value of 0.13 and a coefficient decrement value of 0.002 in hybrid learning consisting of self-learning with a coefficient multiplier value of 1 and a coefficient decrement value of 0.01, and learning with a teacher using a vector quantization algorithm with a learning constant value of 0.15 and a decrement constant value of 0.003. The function of changing the learning coefficient is linear decrease, the classification dataset contains 400 records.