The paper is about processing of biomedical data. It were used 13 methods of machine learning (Naive Bayes classifier for multivariate Bernoulli models, A decision tree classifier, An extremely randomized tree classifier, Classifier implementing the k nearest neighbors vote, Linear Discriminant Analysis, Linear Support Vector Classification, Logistic Regression, Nearest centroid classifier, A random forest classifier, Classifier using Ridge regression, Ridge classifier with built in cross validation, Gaussian Mixture Models, Support Vector Machines) and one method of deep learning (Multiplayer Perception). A discrete wavelet transform was used to extract of biometric features. Haar wavelets, Daubechi wavelets, Symlets, Coiflets, Biorthogonal, Reverse biorthogonal, Discrete Meyer (FIR Approximation) were used. The influence of Electrocardiorams (ECG) recording time on the accuracy of biometric identification and diagnosis of cardiovascular diseases was studied. It was found that the best methods of classification are: Multiplayer Perception, An extremely randomized tree classifier, Classifier implementing the k nearest neighbors vote and Logistic Regression aka logit MaxEnt classifier. Wavelet family doesn't affect significantly on accuracy of recognition. With increasing registration time, accuracy increases .

The paper is about the problem of class imbalance in the diagnosis of diseases of the cardiovascular system using recognition of electrocardiograms. Under researching two oversampling approaches were compared. Complete cardiocycles (600 points) were used as features. In the first case, the bootstrap method was used. For recognition, a Multylayer Perceptron neural network was used. To solve the problem, significant computational resources and high costs of computer time were required. In the second case, cardiocycles were converted into images oversampled by augmentation. The calculation time was reduced from two and a half hours to 15 minutes. In the third case, both approaches were combined, which reduced the computation time to three minutes. In all three cases, recognition accuracy exceeded 97%.