LHOM-Based ECG Signal Classification

This project focuses on classifying ECG signals (Normal vs Diseased) using Local Higher Order Moments (LHOM) as statistical features. It combines signal processing, feature engineering, and machine learning (SVM, KNN, Random Forest) in MATLAB.

Project Objectives

• Simulate or load ECG signals (normal & diseased)
• Extract Local Higher Order Moments (up to 100th order)
• Normalize and log-transform moment features
• Save features to CSV for analysis
• Train ML models (SVM, KNN, Random Forest)
• Analyze performance and overfitting behavior

Folder Structure

LHOM-ecg-classification

├── ecg_normal1.mat # Sample normal ECG signal

├── ecg_disease1.mat # Sample diseased ECG signal

├── extract_lhom_features.m # Main script for feature extraction

├── lhom_features_10000.csv # Saved 10000 samples with LHOM features and labels

├── classify_models.m # SVM, KNN, Random Forest classifiers

├── overfitting_plot.m # Overfitting detection and plotting

└── README.md # Project documentation

Features Implemented:

1. ECG Signal Preprocessing

• Loading .mat files of ECG signals
• Signal repetition to generate 5000 samples/class (10,000 total)
• Z-score normalization of each segment

2. LHOM Feature Extraction

• Moments computed from 1st to 100th order
• Logarithmic transformation using log10(abs(moment) + eps)
• Separate LHOM profile for normal vs diseased ECG

3. CSV Export

• 100 LHOM features per sample
• Final dataset: 10000 x 101 (last column = label)
• Saved as lhom_features_10000.csv

4. Machine Learning Classification

• SVM with RBF kernel
• KNN with k = 5
• Random Forest with 100 trees
• Train-test split (70/30)
• Accuracy and confusion matrix for each

5. Overfitting Visualization

• Varying training sizes: 10% to 90%
• Accuracy plots: Train vs Test
• Helps detect overfitting behavior for all models

6. Sample Output

SVM Accuracy : 96.20% KNN Accuracy : 94.85% Random Forest Acc : 97.40%

7. Overfitting Plot

• SVM, KNN, and RF accuracies shown against increasing train sizes
• Helps evaluate model robustness and generalization

8. Requirements

• MATLAB R2021a or later
• Signal Processing Toolbox (optional but helpful)

9. Future Scope

• Apply PCA or t-SNE on LHOM features for visualization
• Use raw ECG input with CNN or RNN models for temporal classification
• Extend dataset with real ECG samples (e.g., MIT-BIH)
• Explore hybrid feature sets: LHOM + Wavelet + Entropy

10. Citation

If this project helped you, please consider citing it or starring the repository ⭐.