📝 FedOps Fitbit Health Pipeline

Baseline

- Baseline
    - client_main.py
    - client_mananger_main.py
    - server_main.py
    - models.py
    - data_preparation.py
    - requirements.txt (for server)
    - conf
        - config.yaml
    - **xai_utils.py**

Step

First，Start by cloning the FedOps（Place it cloning in the local location）

git clone https://github.com/gachon-CCLab/FedOps.git \
&& mv "FedOps/silo/examples/torch/Wearable(FitBit+XAI)" . \
&& rm -rf FedOps

1. Create Task: The task name is required (e.g., fitbitxai). Since this instance belongs to a general machine learning or deep learning task, select AI. Keep all subsequent options as default, and finally, choose FedAvg as the federated aggregation strategy.

   

   

   Title: Task name, for example fitbitxai.

Model Type: Select AI (for general machine learning or deep learning tasks).

XAI: Select Enabled.

Basic Training Parameters and FedAVG Parameters can be left as default.

Dataset Parameters (Dataset and Model Settings)

Parameter	Example	Description
Dataset Name	fitbitxai	Specifies the dataset used for training.
Model Name	models.SleepLSTM	Specifies the model file (e.g., a custom SleepLSTM model).

After confirming all parameters are correct, click the CREATE button at the bottom to generate the task instance. The new task will then appear in the task list.

1. Enter the server managent of the created task.

2. In Server Management, configure Resource Scaling (the default values are CPU: 1 and Memory: 2 Gi, so modify them if necessary).

   Then, click Create Scalable Server to create the server pod. Once created, this dashboard will show pod and PVC status as in the image above.

   （ {“replicas”:1,”ready_replicas”:1,”available_replicas”:1} is normal status）

3. Enter the server managent of the created task.

4. In Server Management, configure Resource Scaling (the default values are CPU: 1 and Memory: 2 Gi, so modify them if necessary).

   Then, click Create Scalable Server to create the server pod. Once created, this dashboard will show pod and PVC status as in the image above.

   （ {“replicas”:1,”ready_replicas”:1,”available_replicas”:1} is normal status）

   

   

5. To properly load the Fitbit Sleep and Activity Dataset, you need to install the kagglehub library.

   As shown in the image, you can do this by running the following command in the Execute Command section of your server interface:

   pip install kagglehub
   

   

6. When editing or replacing files inside the Pod:

   At the top of the File Browser, enter the path /app/code/ and click Browse to confirm the file directory.

   In the File Content section on the right, type the full file path and click Load for each of the following files:

   • /app/code/models.py
   • /app/code/data_preparation.py
   • /app/code/server_main.py
   • /app/code/conf/config.yaml

   Then, paste the new content you’ve prepared locally into the editor on the right and click Save File to apply the changes.

   

7. Click Set Start Command to prepare the command for running the FL server.

   (Although you can also start the server by clicking Start FL Server, it will only run the server without saving logs. Therefore, it is recommended to use Set Start Command to review and confirm the command before execution.)

   Once the command is ready, click Execute to run it.

   Then, click Check Process to verify that the FL server process is running.

   

8. Run the clients.
   • Run client_main.py and client_manager_main.py
   • Then, in the terminal to confirm whether it runs correctly.

   

9. The monitoring page can confirm the global results

10. XAI result

    

    Two interpretability methods, Integrated Gradients (IG) and LIME (Local Interpretable Model-agnostic Explanations), were applied to explain the feature importance distribution of the federated wearable device model in short-term psychological state prediction tasks. The input features include Steps, Calories, Average Heart Rate (AvgHR), and Stress.

(a) Integrated Gradients (IG)

(b) LIME (Local Interpretable Model-agnostic Explanations)

Interpretation Method	Strength	Finding	Conclusion
Integrated Gradients (IG)	Captures global gradient distribution	Highlights Stress and AvgHR features in the 3–4 h interval	The model globally focuses on physiological stress evolution
LIME	Captures local sensitive features	Strongest Stress response observed in the 3–4 h interval	Local explanations validate the consistency of global patterns