Prediction of heart arrhythmia onset in patients with ICD using machine learning algorithms.

PURPOSE: The current study uses machine learning algorithms to predict the onset of cardiac arrhythmia in patients with implantable cardioverter-defibrillators (ICD). We aimed to check how far in advance the onset of arrhythmia can be predicted. MATERIALS AND METHODS: The input data consisted of 176 signals of R-R intervals from 28 patients with ICD, recorded both during a normal heartbeat and before the onset of arrhythmia. For every signal, we generated 42 descriptors with different signal analysis methods. Then, we identified relevant descriptors using the Boruta algorithm. Then, we used machine learning methods, such as Random Forest, AdaBoost, XGBoost, LASSO, and SVM, and focused on the best results obtained by Random Forest. Cross-validation was used to assess the accuracy of predictions. RESULTS: To check the nature of the signal, we trained the classifiers on the selected temporary windows. The best obtained AUC was 0.75. We have also presented risk groups. CONCLUSION: We found that the signal (information about arrhythmia) already appears for at least 1,000 R-R intervals before the onset of arrhythmia and gets stronger with the decreasing time to the onset.