"# Soldier Activity & Anomaly Detection"

Anomaly Detection and Activity Recognition for Soldier Welfare

Project Status: Completed

This project builds a machine learning system to monitor a soldier's well-being and tactical situation using data from a smartwatch. The final model can successfully classify 3 normal activities and 4 distinct anomalies, including critical health and fall events.

---

1. Project Goal

The system is designed to fulfill two primary functions:

• Soldier Welfare: Provide real-time alerts by detecting anomalous events like falls or health crises.
• Tactical Awareness: Provide a high-level overview of a soldier's current activity state (e.g., "Resting" or "Running").

---

2. The Data

The project uses a simulated smartwatch dataset (data/simulated_smartwatch_data.csv). This was necessary because public datasets (like UCI-HAR) do not contain anomaly data or heart rate features.

The simulated dataset contains 4 raw features and 7 specific labels:

Features (The "Inputs"):

• heart_rate
• accel_x (X-axis acceleration)
• accel_y (Y-axis acceleration)
• accel_z (Z-axis acceleration)

Labels (The "Outputs"):

1. Normal_Resting: Low heart rate, low movement.
2. Normal_Walking: Elevated heart rate, moderate rhythmic movement.
3. Normal_Running: High heart rate, high-impact movement.
4. Anomaly_Fall: High, erratic acceleration.
5. Anomaly_HealthEvent: Critically high heart rate without movement.
6. Anomaly_SensorFailure: All sensors report an error state (-1).
7. Anomaly_DataLoss: All sensors report missing data (0).

---

3. The Model

The final model is a Random Forest Classifier trained on the 7 labels.

• Model File: models/soldier_smartwatch_model.joblib
• Training Notebook: notebooks/03-Smartwatch-Model.ipynb
• Performance: The model achieved ~99-100% accuracy on the test set, successfully distinguishing all 7 normal and anomalous classes.

---

4. How to Use This Project

1. Environment: The project requires the Conda environment defined in soldier_env.
2. Generate Data: Run python src/generate_smartwatch_data.py to create the dataset.
3. Train Model: Open and run the notebooks/03-Smartwatch-Model.ipynb notebook. This will train the model and save it to the models/ folder.