Table 9. Comparison of our results with recent works undertaken towards multi-class respiratory disease classification.

Authors and Year	Dataset used	Features / Input to model	Proposed Model(s)	Sensitivity	Specificity	ICBHI Score
[25]	ICBHI Dataset with healthy and two classes (Chronic and Non-Chronic)	MFCCs combined with their first-order derivative	LSTM	0.98	0.82	0.90
[30]	ICBHI Dataset with CNN VAE generated synthetic samples of healthy and five disease classes (Bronchiectasis, Bronchiolitis, COPD, Pneumonia, URTI)	Mel Spectrograms of respiratory sounds	CNN	0.99	0.99	0.99
[47]	ICBHI Dataset with augmented samples of two classes (COPD and Non-COPD)	MFCCs	CNN	0.92	0.92	0.92
[48]	King Abdullah University Hospital + ICBHI Database with six classes (Normal, COPD, BRON, Pneumonia, Asthma, heart failure)	Entropy-based features	Boosted Decision Trees	0.95	0.99	0.97
Our Study (2021)	ICBHI dataset with VAE-generated synthetic samples of healthy and six disease classes (Pneumonia, LRTI, URTI, Bronchiectasis, Bronchiolitis, COPD)	MFCCs of respiratory sound segments	MLP	0.97	0.51	0.74
			CNN	0.96	0.62	0.79
			LSTM	0.92	0.41	0.67
			RESNET-50	0.98	0.71	0.85
			EFFICIENT NET B0	0.96	0.56	0.76