Algorithm 1. CNN-AE pseudo-code
Input: dataset $D=\{D_{\mathit{recovered}}\cup D_{\mathit{deceased}}$}, training epochs N, batch size B, number of folds K
	// Auto-encoders initialization
1:	Create 10 autoencoders with initial random parameters: {$A{E}_1,\cdots ,A{E}_{10}\}$
	// Autoencoders training
2:	Partition samples in $D_{\mathit{recovered}}$ to 10 subsets:$\{g_1,\cdots ,g_{10}\}$
3:	For i = 1:10
4:	Train $A{E}_i$ on $D_{\mathit{recovered}}-g_i$ and perform validation on $g_i$
	// Augmented data generation
5:	$A=[]$
6:	For i = 1:10
7:	$a_i=A{E}_i(D_{\mathit{deceased}})$
8:	A $=A\cup a_i$
9:	$D_{\mathit{augmented}}=D\cup A$
10:	Create CNN $C$ with initial random parameters
11:	// K-Fold cross validation Partition $D_{\mathit{augmented}}$ to 90% training set $D_{\mathit{train}}$ and 10% test set $D_{\mathit{test}}$
12:	Partition $D_{\mathit{train}}$ to K subsets $\left\{F_1,,,,,\cdots ,,,,,F_K\right\}$
13:	For k = 1:K
14:	$D_{\mathit{train}}=D_{\mathit{augmented}}-F_K$
15:	$D_{\mathit{valid}}=F_K$
16:	For e = 1:N
17:	$batc{h}_t$ = sample_batch($D_{\mathit{augmented}},B$)
18:	CNN.train($batc{h}_t$)
19:	$batc{h}_v$ = sample_batch($D_{\mathit{valid}},B)$
20:	CNN.validate($batc{h}_v$)
21:	CNN.test($D_{\mathit{test}}$)
22:	Return CNN