International Journal of Science, Engineering and TechnologyAuthors: Hrishikesh Shanbhag, Dr. Sonal Ayare
Abstract: Respiratory diseases, notably asthma, pneumonia, and chronic obstructive pulmonary disease (COPD), pose sig- nificant challenges globally. Conventional diagnostic techniques are often resource intensive, invasive, and not scalable for rapid screening. This paper presents a comprehensive deep learning framework using Mel-Frequency Cepstral Coefficients (MFCC) and a hybrid Gated Recurrent Unit – Convolutional Neural Network (GRU-CNN) architecture to detect and clas- sify respiratory diseases from audio samples. We elaborate on data preprocessing, feature extraction, model design, training strategy, and rigorous evaluation. Our approach achieves an overall accuracy above 91%, demonstrating robustness across multiple disease classes. The paper discusses technical insights, challenges in real-world adaptation, comparative analysis with existing methods, and future work to further enhance clinical applicability. Additionally, we explore the clinical implications and provide detailed implementation guidelines for healthcare practitioners and researchers.
