<![CDATA[Chronic obstructive pulmonary disease (COPD) is associated with persistent airflow obstruction that may produce measurable changes in respiratory sounds. Automated lung-sound analysis could support objective respiratory assessment, although its performance and generalizability require careful evaluation. This study developed and internally evaluated a binary COPD classification pipeline combining Mel Frequency Cepstral Coefficient (MFCC) features with a Multi-Layer Perceptron (MLP) artificial neural network. Lung-sound recordings were obtained from the publicly available ICBHI 2017 Respiratory Sound Database. After eligibility assessment, signal-quality screening, and segmentation, 1,000 two-second respiratory-sound segments from 90 subjects were retained, comprising 425 COPD and 575 normal segments. Data were partitioned at the subject level into training, validation, and held-out test subsets in an approximate ratio of 64:16:20, ensuring that segments from the same participant did not appear in different subsets. Each segment was represented by a 26-dimensional feature vector derived from the mean and standard deviation of 13 MFCC coefficients. The classifier consisted of 26 input units, two hidden layers with 64 and 32 rectified linear units, a dropout rate of 0.20, and one sigmoid output unit. The model was trained using Adam, inverse-frequency class weighting, L2 regularization, and validation-based early stopping. On the held-out test set of 200 segments, the model achieved an accuracy of 93.50%, precision of 90.91%, sensitivity of 94.12%, specificity of 93.04%, and an F1-score of 92.49%. These findings demonstrate the internal feasibility of an MFCC–MLP pipeline for distinguishing COPD from normal lung-sound segments within the study dataset. However, the results do not establish clinical diagnostic validity or deployment readiness. Independent patient-level, multicenter, prospective, and device-specific validation is required before the method can be considered for clinical decision support or portable health-care applications.]]>
