Anomaly Detection project in ECG signals

The goal of this project is to detect irregular heart rhythms from ECG signals. Since irregular heart rhythms usually constitute a minority compared to normal ones, we are solving an anomaly detection problem. This is a challenging task for machine learning models, but if it works well, it enables expert technicians and doctors to detect fastly cardiac arrhythmias in patients avoiding any manual and repetitive work.

The article with the explanations is How to deploy your ML model using DagsHub+MLflow+Lambda AWS

Tools used in the project

• DVC
• DagsHub
• MLflow
• AWS Lambda

Project Structure

• src: contains the scripts to train and evaluate the model
• ecg_data: contains the data
• Autoencoder: save the model

Dataset used

ECG5000 is the dataset analyzed in this project. You can find the dataset here. It contains 5000 heartbeats randomly selected from a patient that had heart failure. It was originally found in PhysioNet, a repository of medical data available for researchers, and the class values, that indicate if the heart rhythm is anomalous or not, were obtained by automated annotation.

Set up the environment

1. Clone the repository

git clone --branch branch_2 https://dagshub.com/eugenia.anello/anomaly_detection_ecg.git

2. Create a virtual environment in Python

3. Install the requirements

pip install -r requirements.txt

4. Pull the data

dvc pull

Run experiment

cd src
python train.py