Halo Collar Activity Recognition

GPS and sensor-based machine learning model for canine activity classification

Project Overview

Engineered a GPS and sensor-based machine learning model for canine activity classification using Python and scikit-learn, achieving 89% accuracy across 8 distinct behavioral patterns. This project was developed in Fall 2022 as part of a partnership with PAWS LLC, processing real-time sensor data from 500+ collar devices.

Key Achievements

  • 89% accuracy in classifying 8 distinct canine behavioral patterns
  • Real-time processing of sensor data from 500+ collar devices
  • Partnership with PAWS LLC for production deployment
  • Comprehensive data pipeline for GPS and accelerometer sensor fusion

Technical Implementation

Machine Learning Pipeline

  • Algorithm: Custom ensemble model using scikit-learn
  • Data Processing: Real-time sensor fusion of GPS coordinates and accelerometer data
  • Feature Engineering: Time-series analysis of movement patterns, GPS trajectory analysis
  • Validation: Cross-validation with stratified sampling across different dog breeds and sizes

Data Sources

  • GPS coordinates with timestamp data
  • 3-axis accelerometer readings
  • Collar orientation sensors
  • Environmental context data

Behavioral Pattern Classification

The model successfully distinguishes between:

  1. Walking/Running
  2. Playing
  3. Resting/Sleeping
  4. Eating/Drinking
  5. Barking
  6. Scratching
  7. Exploratory behavior
  8. Aggressive behavior

Impact and Applications

This project demonstrates the practical application of machine learning in IoT devices for pet monitoring and health tracking. The system provides valuable insights for pet owners and veterinarians about:

  • Daily activity levels and exercise patterns
  • Behavioral changes that might indicate health issues
  • Sleep quality and rest periods
  • Social interaction patterns with other animals

Technologies Used

  • Python for data processing and model development
  • scikit-learn for machine learning algorithms
  • pandas/numpy for data manipulation
  • GPS and accelerometer sensor integration
  • Real-time data processing pipelines

The successful deployment across 500+ devices validates the model’s robustness and scalability for commercial IoT applications in the pet care industry.