HYBRID CNN WITH ATTENTION-BASED FEATURE FUSION FOR LUNG DISEASE DETECTION FROM RESPIRATORY SOUNDS USING CLASS-BALANCED OPTIMIZATION

Lung disease detection using respiratory sounds involves analyzing audio recordings of breath sounds to identify abnormalities associated with conditions such as Chronic Obstructive Pulmonary Disease (COPD), pneumonia, and asthma. However, accurate detection remains challenging due to class imbalance in datasets and the difficulty of extracting discriminative features from complex audio signals. This study proposes a novel Hybrid Convolutional Neural Network with Attention-Based Feature Fusion (HCNN-AFF) to enhance detection accuracy. The model integrates spectrogram-based inputs with Mel-Frequency Cepstral Coefficients (MFCCs), effectively capturing both time-frequency and cepstral information for richer feature representation. To address class imbalance, a combination of data augmentation and a class-balanced loss function is employed, enabling improved learning from minority class samples. Furthermore, the model is optimized using the AdamW optimizer enhanced with a LookAhead mechanism (AWLA), promoting better convergence and generalization. Experimental results demonstrate that the proposed HCNN-AFF method outperforms existing approaches in respiratory sound analysis, achieving superior accuracy and robustness in lung disease detection.