Lightweight MFCC–SVM recognition of wheeze events in pediatric respiratory sounds: inter-patient evaluation on SPRSound and a software deployment proof-of-concept

Pediatric wheeze recognition can support the automated analysis of respiratory sounds and reduce dependence on subjective auscultation. This study develops and evaluates a lightweight machine-learning approach for event-level wheeze/no-wheeze recognition using the publicly available SPRSound pediatric respiratory sound database. Respiratory events were extracted exclusively from the expert-annotated temporal boundaries provided with the dataset, without automatic event segmentation. A total of 9,089 annotated respiratory events were represented using Mel-frequency cepstral coefficients (MFCCs), delta, and delta-delta features summarized into a fixed 120-dimensional vector. Logistic Regression, Linear Support Vector Machine, Random Forest, and Support Vector Machine with a radial basis function kernel (SVM-RBF) were evaluated under the same acoustic-feature pipeline. The operating threshold of the SVM-RBF classifier was optimized using the INTRA-2022 partition and subsequently applied to the inter-patient INTER-2022 partition. At this operating point, the model achieved a sensitivity of 0.935 (95% CI 0.901–0.957), specificity of 0.777 (95% CI 0.752–0.801), ROC-AUC of 0.930, PR-AUC of 0.772, and F1-score of 0.679; the positive predictive value was 0.534, reflecting a substantial false-positive burden. As a secondary software-level contribution, the trained preprocessing and classification components were integrated into an Android–Flask client–server workflow for audio submission, server-side feature extraction, inference, and result visualization. The software workflow was functionally implemented but was not quantitatively validated using respiratory recordings acquired through a smartphone microphone. Accordingly, the reported performance characterizes event-level classification under SPRSound conditions and should not be interpreted as mobile-acquisition, diagnostic, or clinical validation.