An augmented ECG data based classification for arrhythmia using optimal feature set.

Area of Science:

• Cardiology
• Biomedical Engineering
• Machine Learning

Background:

• Electrocardiogram (ECG) is vital for diagnosing heart conditions like arrhythmia.
• Prompt arrhythmia detection is critical for preventing adverse events during monitoring.
• Analyzing complex ECG data with machine learning presents significant challenges.

Purpose of the Study:

• To develop an Intelligent Arrhythmia Classification System.
• To enhance diagnostic accuracy for arrhythmias.
• To maintain a lower computational cost in ECG analysis.

Main Methods:

• Developed a preprocessing pipeline incorporating domain knowledge and low-complexity methods.
• Utilized Multilayer Perceptron (MLP) for feature learning and classification.
• Employed Synthetic Minority Over-sampling Technique (SMOTE) to address class imbalance in ECG data.
• Implemented a scalable, real-time multiclass classification system using Kafka and Spark on a three-node cluster.

Main Results:

• Achieved an overall accuracy of 96.4% on the MIT-BIH dataset.
• Demonstrated high performance for Supra-Ventricular Ectopic Beat (SVEB) detection (95.2% positive predictive value, 95.3% sensitivity, 95.24% F1-score).
• Showcased strong results for Fusion Beat (F) classification (91.6% positive predictive value, 91.4% sensitivity, 91.49% F1-score).

Conclusions:

• The developed system outperforms previous methods in arrhythmia classification.
• Achieved superior results with lower computational complexity.
• Effectively handles datasets with limited abnormal beat samples.