A new study compares six deep learning architectures, two Foundation Models (FM), and statistical baselines for multi-horizon behavioral forecasting using mobile health data. The research found that no single architecture consistently outperformed others, with PatchTST leading among trained models, while the FM TimesFM showed strong zero-shot performance, especially in low-data scenarios. Participant-level fine-tuning significantly improved forecasting accuracy, reducing RMSE by 16-60%, with sleep data benefiting the most. AI
IMPACT Provides practical guidance on selecting AI architectures and personalization strategies for mobile health forecasting.
RANK_REASON The cluster contains an academic paper detailing a comparative study of deep learning architectures for a specific application.
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
