The life of modern society is impossible without trust. To ensure trust in the digital world, various encryption algorithms and password policies are used. Passwords are used in a variety of applications from banking applications to email. The advantages of passwords include ease of use and widespread distribution. Forgotten password can be restored or changed. Password weaknesses are largely related to the human factor. Many users use passwords such as "1234" or "qwerty," and they are also willing to share passwords with friends and colleagues. Vulnerabilities are also associated with software and hardware manufacturers. Many Wi-Fi routers preset very simple passwords, which many users leave unchanged. There are questions for manufacturers of mobile applications. Due to the imperfection of their software, personal data of users often leak. Due to the prevalence of social networks, new authentication methods have appeared. On many websites, you can use accounts from Facebook or Gmail.com for authentication. If hackers manage to break into large IT vendors, then millions of accounts will be leaked. Many common password problems can be overcome with biometric identification. In particular, biometric data are very difficult to fake; they usually do not change over time. Widespread methods of biometric identification, such as fingerprinting, retina recognition, and voice recognition have various vulnerabilities unfortunately.

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%.