Mining Massive E-Health Data Streams for IoMT Enabled Healthcare Systems

Area of Science:

• Data Science
• Machine Learning
• Medical Informatics

Background:

• The Internet-of-Medical-Things (IoMT) generates vast data streams, posing challenges for real-time analysis.
• Concept drift, a change in data distribution over time, complicates processing, especially in healthcare where sensor roles may shift.
• Class imbalance in medical data further exacerbates the difficulty of accurate concept drift detection.

Purpose of the Study:

• To propose an Enhanced Reactive Drift Detection Method (ERDDM) to address concept drift in data streams with class imbalance.
• To systematically generate strategies for handling concept drift and class imbalance simultaneously.
• To evaluate the performance of ERDDM against existing methods.

Main Methods:

• Developed the Enhanced Reactive Drift Detection Method (ERDDM).
• Conducted comparative experiments using the Massive Online Analysis (MOA) framework.
• Utilized 48 synthetic datasets designed to simulate data stream characteristics and class imbalance.

Main Results:

• ERDDM demonstrates superior performance compared to three contemporary techniques across various metrics.
• The proposed method effectively handles both abrupt and gradual concept drifts.
• ERDDM shows significant improvements in prediction error, drift detection delay, and latency, particularly with imbalanced data.

Conclusions:

• ERDDM offers a robust solution for concept drift detection in imbalanced data streams.
• The method is well-suited for applications within the Internet-of-Medical-Things (IoMT) domain.
• ERDDM outperforms existing benchmarks, providing a more reliable approach for real-time medical data analysis.