Contextual recommendation modeling in eCoaching with machine learning, X-AI, and semantic ontology

Area of Science:

• Sports Science
• Computer Science
• Environmental Science

Background:

• Physical activity is crucial for health but often limited by adverse weather conditions.
• Existing solutions lack personalized, real-time guidance for weather-dependent exercise.
• Developing adaptive eCoaching systems is essential for promoting consistent physical activity.

Purpose of the Study:

• To develop an automated eCoaching system for personalized physical activity recommendations.
• To integrate real-time weather data with an Ontology framework for semantic representation.
• To evaluate the system's accuracy and interpretability in diverse weather scenarios.

Main Methods:

• Collected 18 months of weather data from thirteen cities in southern Norway.
• Developed algorithms for data annotation, processing, classification, and rule-based recommendations.
• Employed decision tree classification and Local Model-Agnostic Interpretable Explanations (LIME) for model interpretation.

Main Results:

• The decision tree classifier achieved a 99.1% accuracy in classifying activity recommendations.
• Ontology model verification confirmed reliable semantic representation and efficient rule-based modeling.
• System test cases demonstrated accurate and contextually relevant eCoaching recommendations across various weather conditions.

Conclusions:

• The automated eCoaching system effectively promotes continuous physical activity irrespective of weather.
• Personalized, real-time recommendations enhance user engagement and adherence to exercise routines.
• The integration of weather data, Ontology, and machine learning offers a robust solution for adaptive eCoaching.