Automatic Evaluation of Heart Condition According to the Sounds Emitted and Implementing Six Classification Methods

Area of Science:

• Cardiology and Machine Learning
• Biomedical Signal Processing

Background:

• Heart disease is a leading global cause of mortality, necessitating advanced diagnostic tools.
• Accurate classification of heart sounds is crucial for early detection and management of cardiovascular conditions.

Purpose of the Study:

• To implement, compare, and evaluate machine learning algorithms for classifying normal and abnormal heart sounds.
• To assess the effectiveness of different feature extraction techniques and data preprocessing methods.

Main Methods:

• Extracted 52 features from normal, murmur, and extra systolic heart sounds, including statistical features, Linear Predictive Coding (LPC), and Mel-Frequency Cepstral Coefficients (MFCC).
• Applied six machine learning classifiers: k-Nearest Neighbors, Naive Bayes, Decision Trees, Logistic Regression, Support Vector Machine, and Artificial Neural Networks.
• Evaluated model performance using accuracy, specificity, sensitivity, ROC curve, precision, and F1-score on normalized, standardized, and non-normalized datasets.

Main Results:

• Logistic regression on standardized data achieved a specificity of 0.7500 and ROC curve of 0.8405.
• Logistic regression on normalized data yielded a specificity of 0.7083 and ROC curve of 0.8407.
• Support Vector Machine with a linear kernel on non-normalized data achieved a specificity of 0.6842 and ROC curve of 0.7703.

Conclusions:

• Machine learning models demonstrate significant potential for accurate heart sound classification.
• Logistic regression and Support Vector Machine algorithms are effective tools for computer-assisted diagnosis of heart conditions.
• Feature engineering and data preprocessing significantly impact the performance of heart sound classification models.