Heart Disease Detection using Audio Dataset

Heart disease detection plays a critical role in the early diagnosis and management of cardiovascular diseases. This study explores the application of audio-based methods for heart disease detection using a dataset of heart sounds. Sourced from Kaggle, the dataset consists of labeled heart sound recordings categorized as either "Healthy" or "Unhealthy." The proposed approach employs a range of deep learning models, including Convolutional Neural Networks (CNN), Recurrent Convolutional Neural Networks (RCNN), ResNet50, VGG16, and MobileNet, for effective classification of heart sounds.

For feature extraction, we utilize Mel-frequency cepstral coefficients (MFCC), a robust technique widely used in speech and audio processing for capturing the spectral characteristics of heart sounds. To improve model robustness and generalization, we enhance the dataset with techniques such as noise augmentation, pitch shifting, Zero-Crossing Rate (ZCR), and Root Mean Square Error (RMSE). These methods help address variations in heart sound recordings caused by environmental factors or physiological changes.The CNN and RCNN models, in combination with advanced architectures like ResNet50, VGG16, and MobileNet, are trained to classify heart sounds into "Healthy" and "Unhealthy" categories.

Keywords: Heart disease detection, audio dataset, Kaggle, heart sound classification, CNN, RCNN, ResNet50, VGG16, MobileNet, MFCC, noise augmentation, pitch shifting, ZCR, RMSE, feature extraction, deep learning, machine learning, cardiovascular disease, audio-based diagnosis.