Automated Detection of Heart Disease Using PhysioNet ECG Signals and Machine Learning Models

Heart diseases remain one of the leading causes of death and disability worldwide, even with the immense development of medical diagnosis and treatment. It is very important to diagnose cardiovascular conditions early using non-invasive methods, like the analysis of electrocardiogram (ECG) signals, for better prognosis and the reduction of mortality rates. In this regard, an automated system of heart disease detection is presented, based on the utilization of ECG data from the PhysioNet database in collaboration with a newly developed machine learning-based framework. It combines convolutional neural networks (CNNs) for automatic feature extraction and support vector machines (SVMs) for classification to take collective advantage of the strengths of both models. Besides this, the hybrid model utilizes temporal features like QRS-duration and RR intervals, as well as frequency-domain features derived using Fast Fourier Transform (FFT) for its enhanced predictive power. Results yielded a classification accuracy of 100% on both the training and validation sets, thereby showing the efficacy of the model proposed in this paper. However, the major limitation that was noticed from this model is the clear indication of overfitting-the model does not generalize well on the unseen test data. Future work for us includes working on cross-validation, augmentation techniques, and other regularization methods such as L2 regularization and dropout to generalize better and avoid overfitting. Despite such limitations, our study demonstrates that a hybrid approach using CNN and SVM using both time- and frequency-domain features has great potential in the development of real-time non-invasive diagnostic tools for the detection of heart disease. Once this method is further refined and validated in a larger patient cohort, it may easily be integrated into clinical settings for enabling early interventions and improving patient outcomes.