Developing Machine Learning and Statistical Tools to Evaluate the Accessibility of Public Health Advice on Infectious Diseases among Vulnerable People

Table 3

Performance of Gaussian Naïve Bayes (GNB) classifiers with different feature sets.

Model	Techniques	Training (5-fold CV)	Testing
		AUC mean (SD)	AUC	Accuracy	Macro F1	Sensitivity	Specificity
1	MLS + POS full (69)	0.971 (0.0212)	0.940	0.921	0.992	0.944	0.8824
2	MLS + POS jointly optimised (6)	0.998 (0.0026)	0.993	0.940	0.943	0.963	0.9118
3	MLS full (26 features)	0.997 (0.003)	1.0	0.966	0.963	1.0	0.9118
4	MLS optimised (2 features)	0.998 (0.004)	1.0	0.943	0.938	1.0	0.8529
5	POS full (46 features)	0.959 (0.0238)	0.907	0.852	0.843	0.889	0.7941
6	POS optimised (8 features)	0.982 (0.0166)	0.968	0.955	0.951	1.0	0.8824
7	MLS + POS separately optimised (10)	1.0 (0)	1.0	0.955	0.951	1.0	0.8824
8	Refined MLS + POS separately optimised (2)	0.995 (0.0079)	0.999	0.989	0.988	0.982	1.0