Table 6.

Accuracy gain comparison when using no SSDL (No MM) vs. MixMatch with the proposed loss balancing correction (MM$+$PBC), and to using MixMatch with no balancing correction (MM) vs. MixMatch with the proposed loss balancing correction (MM$+$PBC). The accuracy gain is evaluated for the tested number of labelled observations (10, 15 and 20). Italic entries correspond to non statistically meaningful gains, after performing a Wilcoxon test, with $p>0.1$.

SSDL	COVID-$19^{-}$	COVID-$19^{+}$	Comparison	$n_l=10$	$n_l=15$	$n_l=20$
				Acc. gain	Acc. gain	Acc. gain
Costa Rican dataset	70%	30%	MM$+$PBC vs. No MM	$+$0.07	$+$0.046	$+$0.018
			MM$+$PBC vs. MM	$+$0.048	$+$0.046	$+$0.036
	80%	20%	MM$+$PBC vs. No MM	$+$0.081	$+$0.06	$+$0.038
			MM$+$PBC vs. MM	$+$0.034	$+$0.038	$+$0.041
Chinese paediatric dataset	70%	30%	MM$+$PBC vs. No MM	$+$0.113	$+$0.115	$+$0.06
			MM$+$PBC vs. MM	$\mathit{+\; 0.023}$	$\mathit{+\; 0.031}$	$\mathit{+\; 0.028}$
	80%	20%	MM$+$PBC vs. No MM	$+$0.09	$+$0.058	$+$0.031
			MM$+$PBC vs. MM	$+$0.1	$+$0.099	$+$0.066
Chest X-ray8 dataset	70%	30%	MM$+$PBC vs. No MM	$+$0.083	$+$0.092	$+$0.087
			MM$+$PBC vs. MM	$\mathit{+\; 0.033}$	$+$0.055	$+$0.057
	80%	20%	MM$+$PBC vs. No MM	$+$0.069	$+$0.096	$+$0.056
			MM$+$PBC vs. MM	$+$0.063	$+$0.0774	$+$0.072
Indiana dataset	70%	30%	MM$+$PBC vs. No MM	$+$0.048	$+$0.063	$+$0.027
			MM$+$PBC vs. MM	$+$0.045	$+$0.082	$+$0.067
	80%	20%	MM$+$PBC vs. No MM	$+$0.014	$+$0.059	$+$0.029
			MM$+$PBC vs. MM	$\mathit{+\; 0.056}$	$+$0.141	$+$0.09