HEARTBEAT ABNORMALITY DETECTOR

Heart beat is a specific time-series waveform, which requires temporal and frequency domain tools, while mantaining the time-series structure.

​

For this analysis, we work on a dataset available in Kaggle, for public domain. We will work first on feature engineering, over the envelope, tempo and spectogram. We then build a classifier, to identify abnormal hearts from healthy ones. We will assess its ROC curve as well as indicators like accuracy and recall.

​

The dataset consists of records from two sources: general public from iSethoscope Pro iPhone and digital stethoscope DigiScope readings from hospitals.

​

For the first part of our analysis, we use four waveforms: one healthy (normal), one containing background noise (artifact) and two abnormalities: extrahls and murmur. You can click and listen to them in the buttons below.