Abstract
An investigation into the use of a unifying Homogeneous Feature Vector Representation (HFVR), to address the challenge of applying machine learning and/or deep learning to heterogeneous data, is presented. To act as a focus, Atrial Fibrillation classification is considered which features both tabular and Electrocardiogram (ECG) time series data. The challenge of constructing HFVRs is the process for selecting features. A mechanism where by this can be achieved, in terms of motifs and discords, with respect to ECG time series data is presented. The presented evaluation demonstrates that more effective AF classification can be achieved using the idea of HFVR than would otherwise be achieved.