- A machine learning project to classify heartbeats from ECG.
- Highest accuracy 97%
- Data from the MIT-BIH heartbeat database: https://physionet.org/physiobank/database/mitdb/
- Uses the wfdb library to help with pre-processing the raw files.
- Each heartbeat is extracted using QRS detection.
- The respective label is found from the .atr file and applied to the heartbeat.
- The heartbeats get edge padded to be 450 samples long. (Sample Rate 360hz)
- The related files are readData.py and filterData.py
A sample of the training data
- CNN's appear to be the most effective way to classify the data.
- Train/Test split of 70/30.
- Built using Keras/TensorFlow.
| Results | Network |
|---|---|
 |
 |
- Expand data set to use more sources with similar classification scheme.
- Apply new ML methods to achieve higher accuracy.
- With more data, potentially see if model can be trained on one data set, then applied to a different one while maintaining accuracy.