This work introduces the Entropy Triangle Method (ETM), a novel machine‑learning framework for predicting cardiac rhythms. The pipeline consists of three stages: feature engineering, entropy‑triangle oversampling, and disease prediction. In the first stage, time‑domain, frequency‑domain and nonlinear descriptors are extracted from 12‑lead ECG recordings. The second stage generates synthetic samples by exploiting the entropy‑triangle inequality in feature space, thereby mitigating the severe class imbalance of non‑sinus rhythms. Finally, classifiers such as support vector machines (SVM) are trained to discriminate among 11 rhythm classes (5 sinus, 6 non‑sinus). Experiments on a public ECG dataset with 10,646 patients demonstrate that the entropy‑triangle‑based SVM achieves over 85% accuracy on non‑sinus rhythm detection, and the "shark scent" oversampling technique yields the highest recall. This is the first report of a machine‑learning approach that predicts non‑sinus rhythms with such high precision, offering a promising tool for early arrhythmia prevention.
Review