Rpeak
Rpeak

ECG Beat Classification Model Training

Train a deep learning model for ECG heartbeat classification using the AAMI 5-class standard and optimized 360Hz sampling.

Model Specifications:

• Sampling Rate: 360 Hz

• Beat Window: 135 samples (375ms)

• Input Shape: [135, 1] tensor for CNN

• Architecture: 4-layer 1D CNN, batch norm, dropout, global pooling, dense layers, softmax output

• Optimizer: Adam (lr=0.001)

• Training Data: 48 MIT-BIH records

• AAMI Classes: Normal, Supraventricular, Ventricular, Fusion, Other

Training Details:

• ~100,000+ labeled heartbeat examples

• Balanced dataset with equal AAMI class representation

• Z-score normalized 375ms beat windows

• 70/15/15% train/validation/test split

AAMI Beat Classification:

• Normal: N, L, R, e, j (sinus beats)

• Supraventricular: A, a, J, S (atrial)

• Ventricular: V, E, r (PVCs)

• Fusion: F (mixed beats)

• Other: Q, /, f, n (artifacts)

Training Progress & Logs

Training logs will appear here when training starts...

Training Process (CNN AAMI-5)

  1. Load 48 MIT-BIH ECG records at 360Hz
  2. Extract 135-sample beat windows around R-peaks (375ms)
  3. Map beat annotations to AAMI 5-class standard
  4. Apply Z-score normalization for training stability
  5. Augment beats for device robustness (noise, baseline, scaling)
  6. Balance dataset across all 5 AAMI arrhythmia classes
  7. Train deep CNN model for 10 epochs with validation
  8. Evaluate performance with class-specific metrics (Precision, Recall, F1)

Note: This model uses 360Hz sampling and 135-sample windows, optimized for real-time ECG analysis and improved device generalization.

Model Inspector

Loading model...