Adopting Trustworthy AI for Sleep Disorder Prediction: Deep Time Series Analysis with Temporal Attention Mechanism and Counterfactual Explanations

Multiple AI models improve prediction reliability: The research combines three different deep learning approaches (TCN, LSTM, and Temporal Fusion Transformer) rather than relying on a single model, which helps ensure more robust and trustworthy sleep disorder predictions across different data patterns.

Explainability is built into the system from the start: The study integrates temporal attention mechanisms and SHAP-based counterfactual explanations directly into the prediction process, meaning doctors and patients can understand why the AI made a specific prediction about sleep disorders—not just what it predicted.

Temporal patterns matter for health predictions: By analyzing how lifestyle and physiological data change over time (rather than just looking at single snapshots), the models can better identify early warning signs of sleep disorders, potentially enabling earlier intervention and treatment.

Transparency supports clinical adoption: By making the AI's decision-making process interpretable through attention mechanisms and counterfactual explanations, healthcare providers are more likely to trust and adopt these predictions in real clinical settings, rather than treating the AI as a "black box."

Personalized insights through counterfactual analysis: SHAP counterfactual explanations can show patients what specific changes in their lifestyle or health metrics would alter the prediction, providing actionable, personalized recommendations for sleep improvement.