Advanced Deep Learning Techniques for Lung Sound Classification: Binary, Multi-Class, and Ensemble Approach

Detecting respiratory diseases, particularly lung disorders is crucial due to their global prevalence. Auscultation by experts has been the standard method for detecting lung diseases. However, the efficiency of diagnosis can be greatly increased by automating this procedure. Using Convolutional Neural Networks (CNN) for binary classification and a CNN, CNN-LSTM hybrid model for multi class classification, this study uses deep learning approaches to automate lung sound classification. To increase classification accuracy, the study extracts important elements from lung sounds such as chromograms, Mel spectrograms and Mel Frequency Cepstral Coefficients (MFCCs). Ensemble learning which combines CNN and CNN-LSTM for multi-class classification, substantially improves the model. With an accuracy of 86.11% the suggested ensemble model surpasses current models in the field. The accuracy of binary classification is 88.38%. This study demonstrates the advantages of combining multiple architectures to improve classification performance and shows how deep learning models can be used to effectively classify lung sounds. The results have significant implications for the early detection of lung diseases.