Page 147 - AI for Good Innovate for Impact

AI for Good Innovate for Impact

(continued)

Item Details
Model Training and 1. Data Sources
Fine-Tuning The AI models are trained on a combination of 4.1-Healthcare
• Structured health system data (e.g., appointment logs, wait times,
patient journey touchpoints)
• Anonymised user interaction data from mobile and web platforms
• Environmental and demographic data, including user location,
internet speed, and device type
• Accessibility interaction logs to improve generative AI features for
underserved populations
• All data is collected and processed in compliance with UAE data
privacy laws and international health data regulations (HIPAA/
GDPR principles).
2. Initial Model Training
• Supervised learning is used for predictive models (e.g., queue/wait
time estimation, appointment load balancing).
• Unsupervised & Semi-Supervised Learning support behavioral
segmentation and clustering to optimize personalized digital
health journeys.
• Large Language Models (LLMs) fine-tuned on healthcare-relevant
Arabic and English corpora power Generative AI modules used in
accessibility and patient assistance.
3. Fine-Tuning Approach
Continuous Learning Loops
• Models are periodically fine-tuned based on new patient flow
patterns, updated scheduling policies, seasonal demand trends,
and feedback from end-users.
Domain-Specific Adaptation
• Pre-trained base models are fine-tuned using health-specific
vocabularies, clinical intent data, and real-world use cases from
Emirates Health Services environments.
Personalization Layer
• Lightweight user context (e.g., prior visits, digital behavior, location,
language preference) is injected into model responses for custom-
ized experience delivery.

Bias Auditing and Model Validation
• Regular audits ensure models perform equitably across different
patient demographics, and validation datasets are refreshed to
match evolving user behavior.

Testbeds or Pilot It can be initiated by logging in to Emirates Health Services website: [1]
Deployments

111