Comparing Binary Classification Results for Dangerous Ventricular Arrhythmias Using 2 and 4 Seconds Short ECG Signal Fragments

The study, conducted by scientists from the Department of Biotechnical Systems of the St. Petersburg State Electrotechnical University with the support of the National Medical Research Center named after V.A. Almazov, is devoted to the binary classification of dangerous ventricular arrhythmias using short 2- and 4-second fragments of ECG signals. The study used two databases of ECG fragments containing spectral descriptions of the signals, and used various classification methods, including nearest neighbor (kNN), linear discriminant analysis (LDA), support vector machine (SVM) and random forest. The results showed that models trained on 4-second data fragments had better metric results in the test compared to models trained on 2-second fragments, with the highest accuracy demonstrated by data from 4-second fragments using 30 harmonics and a frequency step of 0,5 Hz. The study was supported by the Russian Science Foundation grant No. 23-21-00215.