BFRB Detection using Sensor Data

This project focuses on detecting Body-Focused Repetitive Behaviors (BFRB) using sensor data, including IMU, thermal, and time-of-flight sensors. The goal is to classify different types of BFRB gestures to aid in behavioral therapy and monitoring.

Project Structure

cmi-detect-behavior-with-sensor-data/
├── data_preparation_full/      # Full sensor data processing
├── data_preparation_imu_only/  # IMU-only data processing
├── kaggle_evaluation/          # Kaggle competition evaluation code
├── analyze_features.py         # Feature analysis scripts
├── baseline_model.py           # Baseline model implementation
├── bfrb_models.py             # Advanced BFRB detection models
├── data_preparation_steps.py   # Data preprocessing pipeline
├── enhanced_feature_engineering.py  # Advanced feature engineering
├── explore_data.py            # Data exploration scripts
├── feature_engineering.py      # Basic feature engineering
├── requirements.txt            # Python dependencies
├── run_bfrb_pipeline.py        # Main pipeline execution
├── sequence_models.py          # Sequence-based models (LSTM, etc.)
└── train_sequence_models.py    # Training scripts for sequence models

Features

• Multi-sensor data processing: Handles IMU, thermal, and time-of-flight sensor data
• Advanced feature engineering: Time-domain, frequency-domain, and statistical features
• Multiple model architectures:
  • Traditional ML models (XGBoost, LightGBM)
  • Deep learning models (LSTM, CNN, etc.)
  • Ensemble methods
• Kaggle competition ready: Includes evaluation framework

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/cmi-detect-behavior-with-sensor-data.git
   cd cmi-detect-behavior-with-sensor-data

2. Create and activate a virtual environment (recommended):

   python -m venv venv
   source venv/bin/activate  # On Windows: .\venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

Usage

1. Data Preparation:

   python data_preparation_steps.py

2. Feature Engineering:

   python enhanced_feature_engineering.py

3. Train Models:

   • For traditional ML models:

     python run_bfrb_pipeline.py

   • For sequence models:

     python train_sequence_models.py

4. Evaluate Models:

   • Run the Kaggle evaluation framework:

     python kaggle_evaluation/cmi_inference_server.py

Data

The dataset includes:

• Sensor Data:
  • IMU (acc_x, acc_y, acc_z, rot_w, rot_x, rot_y, rot_z)
  • Thermal sensors (thm_1 to thm_5)
  • Time-of-Flight sensors (tof_1 to tof_3)
• Demographics: Age, sex, handedness, height, arm measurements
• Labels: 8 target BFRB gestures and 10 non-target behaviors

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Kaggle competition: CMI Detect Behavior with Sensor Data
• Special thanks to the competition organizers and data providers