A medical device and method for detecting a ventricular arrhythmia event is disclosed. The medical device includes input circuitry configured to receive an electrocardiogram (ECG) signal, processing circuitry coupled to the input circuitry and configured to identify at least one fiducial point of a first heartbeat signature and at least fiducial point of a second heartbeat signature of the ECG signal, and feature extraction circuitry coupled to the processing circuitry. The feature extraction circuitry is configured to determine at least one difference between the at least one fiducial point of the first heartbeat signal and the at least one fiducial point of the second heartbeat signal. Machine learning circuitry is coupled to the feature extraction circuitry and is configured to select a ventricular arrhythmia class based on the at least one difference.