YouTube video by DevisMoriarty

**Abstract:** This paper proposes a novel framework for improving the accuracy of longitudinal prediction of treatment response in digital therapeutics (DTx) addressing depression and anxiety. Leveraging a Multi-Modal Bayesian Network Integration (MMBNI) approach, we combine passively collected behavioral data (sleep patterns, activity levels, mobile usage), patient-reported outcomes (PROs) integrated from DTx platforms, and clinical assessments to create a more robust predictive model.

**Abstract:** Existing methods for evaluating the coherence and fidelity of Explainable AI (XAI) explanations often rely on static datasets and simplistic evaluation metrics, failing to account for the dynamic interplay between model behavior, explanation content, and user interpretability. This paper introduces a novel framework, the Multi-Modal Bayesian Network Inference (MMBNI) approach, to critically assess XAI explanations by constructing probabilistic models incorporating model behavior, explanation characteristics, and human cognitive responses.