Empowering Patient Similarity Networks through Innovative Data-Quality-Aware Federated Profiling.

Area of Science:

• Edge Computing
• Data Quality Assurance
• Machine Learning

Background:

• Continuous patient monitoring generates vast sensory data, necessitating edge processing for efficiency and privacy.
• Data quality (DQ) degradation at the edge, due to sensor issues or transmission problems, can compromise clinical decisions.
• Rapid identification of data quality issues is critical to prevent misinterpretations in patient care.

Purpose of the Study:

• To propose Federated Data Quality Profiling (FDQP) for assessing data quality at edge nodes in patient monitoring systems.
• To develop a formal model for FDQP that captures data quality dimensions and guides node-level data quality assurance.
• To leverage federated learning principles for efficient and privacy-preserving data quality assessment.

Main Methods:

• Developed a formal model for Federated Data Quality Profiling (FDQP) to capture data quality dimensions.
• Employed federated feature selection, ranking features by value, outlier percentage, and missing data percentage.
• Experimented with a fetal dataset distributed across edge nodes to evaluate the FDQP model under various scenarios.

Main Results:

• The proposed FDQP approach demonstrated a significant improvement in data quality at the edge.
• FDQP positively impacted the accuracy of federated patient similarity network (FPSN)-based machine learning models.
• Lightweight profile exchange in FDQP achieved optimal data quality with improved efficiency compared to full data processing.

Conclusions:

• FDQP is an effective method for assessing and improving data quality in edge computing environments for patient monitoring.
• The data-quality-aware federated architecture leveraging FDQP enhances the accuracy of machine learning models.
• The FDQP approach shows potential for application in diverse scenarios beyond patient monitoring.