Enhanced Cough Analysis Using 1-Dimensional CNN Features for Respiratory Health Diagnosis

Cough, a prevalent respiratory symptom, plays a crucial role in disease diagnosis. The accurate differentiation between wet and dry coughs is essential for effective respiratory health assessment. This work presents a robust method for classifying cough audios into wet and dry utilizing one dimensional Convolutional Neural Network (1D-CNN) model as a feature extractor. A hyperparameters tuning approach is introduced which in turn optimizes the CNN model and tunes the size of the pre-last dense layer. Features are extracted from undersampled and oversampled versions of the dataset and are introduced to Random Forest (RF), Support Vector Machine (SVM) and Logistic Regression (LR) classifiers for performance comparison. The research demonstrates that the hyper-tuned feature extractor combined with the RF classification model using the augmented version of the dataset surpasses the other trials by achieving a validation accuracy of 99.29% and ROC AUC of 100%. A remarkable increase of 4 to 16% in the accuracy and of 14% in the ROC AUC is obtained when comparing with the literature. The results indicate the effectiveness of the proposed approach in enhancing the automatic detection of cough types in the healthcare sector. Overall, this work holds great promise for improving healthcare diagnostics and enabling remote disease detection and monitoring.