ECG Classification Models

This repository contains two deep learning models for ECG classification: one using a CNN-LSTM architecture trained on the MIT-BIH Arrhythmia dataset for multi-class classification, and another using CNN and BiLSTM layers on the PTB Diagnostic ECG Database for binary classification of normal and abnormal ECG signals.

1. MIT-BIH ECG Classification using CNN-LSTM Model

Classifies ECG heartbeats using the MIT-BIH Arrhythmia dataset. The model leverages CNN for feature extraction and LSTM for sequence modeling.

Model Architecture:

• CNN Layers: 1D convolutional layers for feature extraction.
• LSTM Layers: Bidirectional LSTM layers to capture temporal dependencies.
• Output: Softmax activation for multi-class classification.

Performance:

• Accuracy: 98.43%
• Macro Avg: 92.75%
• Weighted Avg: 98.40%

2. ECG Classifier using CNN + BiLSTM

Classifies ECG signals as normal or abnormal using the PTB Diagnostic ECG Database (PTBDB). The model combines CNN for feature extraction and BiLSTM for sequence modeling.

Model Architecture:

• CNN Layer: 1D convolutional layer.
• BiLSTM Layers: Four Bidirectional LSTM layers.
• Output: Sigmoid activation for binary classification.

Performance:

• Accuracy: 99.08%
• Macro Avg: 98.85%
• Weighted Avg: 99.08%

Requirements:

• tensorflow, keras, numpy, pandas, scikit-learn, matplotlib