Enhancing YAMNet Model for Lung Sound Classification to Identify Normal and Abnormal Conditions

Healthy lung sounds are produced by airflow during normal breathing. Normal lung sounds lack additional sounds such as rhonchi, wheezing, stridor, or crackles. Abnormal lung sounds result from airflow through an impaired respiratory tract. This research classifies normal and abnormal lung sounds using the YAMNet model, a deep-learning model capable of identifying normal and abnormal lung sounds. The dataset for this research was obtained from Fortis Hospital India and Kaggle. The study involved comprehensive preprocessing of lung sound signals, including sampling at a frequency of 4 kHz, segmenting the lung sound signal for 6 seconds, and smoothing the signal using the Wavelet Smoothing technique, as well as Min-Max Normalization. A 10-fold cross-validation technique was employed, where each iteration used one of the ten parts as the test dataset and the other nine as the training dataset. This model was trained and tested using a 10fold cross-validation technique with an average accuracy of 92.02%. The research yielded accuracy values of 89.81%, precision of 88.53%, recall of 89.74%, and an F1-score of 89.14%.