PCG Sound Quality Classification Based Mel-Frequency Cepstral Coefficients Features and Convolutional Neural Network

Chambi E.M.; Sulla-Espinoza E.; Rendulich J. · Proceedings of the LACCEI international Multi-conference for Engineering, Education and Technology (2023)

Resumen

PCG signals are important to give diagnoses of any disease, specifically the problem of incorrect data collection is addressed, either by an inexperienced doctor or by the patient himself. In this article we seek to classify the signals by their quality labeled as 'Acceptable and Unacceptable' with balanced data and the proposed method-based Mel frequency cepstral coefficients (MFCC) to feature extraction and convolutional neural networks (CNN) for their classification. The results showed us a high performance of our model with 95.3% precision, 95.1% recall, 95.2% F1-score and 95.3% accuracy and it will be compared with already existing works.

Instituciones: Universidad Nacional de San Agustín de Arequipa