Exploring equine behavior: Wearable sensors data and explainable AI for enhanced classification.

Area of Science:

• Animal behavior
• Machine learning
• Wearable technology

Background:

• Advanced monitoring is key for equine welfare and health.
• Wearable sensors capture detailed equine motion data.
• Explainable AI (XAI) is needed to interpret complex behavior models.

Purpose of the Study:

• To integrate wearable sensor data with XAI for equine behavior classification.
• To enhance the interpretability of AI models in equine studies.
• To identify key sensor features for distinguishing equine behaviors.

Main Methods:

• Utilized an open-source dataset of equine behavior from 18 horses.
• Employed machine learning models (Random Forest, KNN, XGBoost) for multi-class classification.
• Applied SHAP (Shapley Additive Explanations) for feature attribution analysis.

Main Results:

• Random Forest achieved 82.3% accuracy in classifying 17 equine behaviors.
• SHAP analysis identified sensor contributions: accelerometer for locomotion, magnetometer for orientation, gyroscope for dynamic movements.
• Specific sensor features were linked to behaviors like galloping, standing, and head-shaking.

Conclusions:

• XAI, specifically SHAP, significantly improves the interpretability of AI models for equine behavior.
• This approach offers actionable insights for real-time monitoring, stress detection, and veterinary interventions.
• The study establishes a new benchmark for explainable AI in equine behavior analysis, enhancing trust and applicability.