Positive and Negative Evidence Accumulation Clustering for Sensor Fusion: An Application to Heartbeat Clustering

Area of Science:

• Data Science
• Machine Learning
• Biomedical Signal Processing

Background:

• Clustering algorithms often struggle with integrating data from multiple sensors.
• Ensemble clustering methods offer a framework for combining multiple data sources.
• The concept of 'negative evidence' in clustering is underexplored.

Purpose of the Study:

• To introduce a novel ensemble clustering algorithm, PN-EAC, designed for multi-sensor data integration.
• To incorporate the concept of negative evidence alongside positive evidence for improved clustering.
• To validate the algorithm's performance in the domain of electrocardiogram (ECG) heartbeat clustering.

Main Methods:

• Developed the PN-EAC algorithm based on the ensemble clustering paradigm.
• Integrated positive evidence (e.g., heartbeat morphology) and negative evidence (e.g., inter-beat distances).
• Applied PN-EAC to heartbeat clustering using data from the MIT-BIH Arrhythmia and INCARTDB databases.

Main Results:

• Achieved a 1.44% error rate on the MIT-BIH Arrhythmia database.
• Obtained 0.601% error on the INCARTDB database using two ECG leads.
• Demonstrated statistically significant improvements with increasing ECG leads (up to 12), reaching 0.338% error.

Conclusions:

• PN-EAC effectively merges data from multiple sensors using ensemble clustering.
• The inclusion of negative evidence enhances clustering performance.
• PN-EAC is the first algorithm capable of simultaneously processing multiple ECG leads for heartbeat clustering.