Principal Component Analysis based on data characteristics for dimensionality reduction of ECG recordings in arrhythmia classification