The goal of this project is to create a program that can predict a user's predicted biometric features and average/max heart rate for a future run based on these route features: Distance, Pace, Elevation gained, Elevation lost. This prediction is made using a machine learning model trained on the user's historical running data through either Garmin Connect or Strava (although Strava provides less features).
Anil Chintapalli and Arjun Rao
- Tensorflow - This machine learning library allows us to train and run models.
- pandas - Library for managing dataframes
- numpy - Library for array manipulation
- Flask - This framework will be used to build the API.
- Streamlit - This framework will be used to build the UI
- Keras - Library for neural network implementation
- Garmin Connect- This API wrapper allows access to a user's Garmin Connect activity data given they provide their associated email, password, start date, and end date.
- SDV - Library to generate synthetic and evaluate the quality of that data
- Scikit Learn - Library for standardization of features before training
- Pytorch - Library for nerual network implementation (preferred over Keras because of warning supression and compatible versions
- Deployment on desktop UI
- Model makes synthetic data from historical running data
- Machine learning model preforms on four inputted individual run parameters
- Outputs future run characteristics for the user (predicted biometric features and average/max heart rate)
1. Obtain and format data
| Status | Date | Description |
|---|---|---|
| Completed | 9/11/2024 | Download personal Garmin connect data to local machine |
| Completed | 9/18/2024 | Format the necessary running activity data as a CSV |
2. Use sdv to generate personal synthetic data
| Status | Date | Description |
|---|---|---|
| Completed | 11/3/2024 | Install sdv |
| Completed | 11/4/2024 | Use sdv to generate 1000 runs for regression |
4. Train/test machine learning model
| Status | Date | Description |
|---|---|---|
| Completed | 12/6/2024 | Neural Network with seperately optimized hyperparameters for best prediction. Solid predictions (within 10 MAE) |
5. API/UI Development
| Status | Date | Description |
|---|---|---|
| Completed | 11/27/2024 | Include way to connect with another user's garmin data (Garmin Connect API Wrapper) |
| Completed | 11/28/2024 | Create UI and API |
| Completed | 12/9/2024 | Integrated machine learning model with the API (didn't work extremely well? Problems with Flask disconnecting) |
| Completed | 12/12/2024 | Workaround that removes API dependancy to have the user's credentials go directly to Garmin Connect from the UI |
6. API/App Development
| Status | Date | Description |
|---|---|---|
| Completed | 1/20/2024 | Finish development of App in Dart (Firebase for account system) |
| In progress | 12/20/2024 | Reinstate backend API/virtual machine for remote usage (especially with full hyperparameter tuning model) |
| In progress | 12/22/2024 | Explore methods for reliable, secure remotely-hosted API |
7. Strava Integration
| Status | Date | Description |
|---|---|---|
| Completed | 2/10/2024 | Integrate PulseRun with Strava (through API support) |