Data Adaptive Stochastic Ensemble Net: Optimizing Infection Predictions for COVID-19 Cluster Analysis.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Epidemiological Modeling

Background:

• Machine learning is crucial in medicine, requiring robust models for accurate AI-driven health predictions.
• Developing effective AI systems for infection prediction necessitates comprehensive datasets and sophisticated modeling techniques.

Purpose of the Study:

• To introduce a novel machine learning model, the Data Adaptive Stochastic Ensemble Network (DASEN), for enhanced infection prediction.
• To improve the robustness of AI-based infection prediction systems using real-world COVID-19 data.

Main Methods:

• Collected a real-world COVID-19 cluster dataset (8,844 cases, 519 clusters) including individual properties and contact relationships.
• Developed DASEN, which optimizes the Dirichlet distribution concentration parameter to dynamically adjust component weights based on data distribution.

Main Results:

• DASEN demonstrated superior robustness across various settings with minimal parameter optimization overhead.
• The model prevented overfitting to majority labels by allowing different components to focus on distinct features.
• Showcased the validity of DASEN in handling data with varying characteristics.

Conclusions:

• DASEN offers a robust and adaptable approach for AI-based infection prediction systems.
• The dynamic weighting mechanism enhances model performance and generalizability in medical applications.
• This study highlights the potential of adaptive ensemble methods in improving healthcare AI.