Algorithm 1. Forward stepwise subset selection.

Variables:

$S_{\text{Candidate}}$—Candidate set

$S_{\text{Target}}$—Target set

$X_{\text{ECG}}$—Single-lead ECG

$F{1}_{\text{ECG}}$—F1 scores for the addition of single-lead ECG

$F{1}_{\text{best},i}$—The best F1 scores in step i

$P_{\text{ECG}}$—The p-value measuring the improvement for the addition of single-lead ECG

$X_{\text{best, }i}$—The best single-lead ECG in step i

Pseudo code:

Initialize

$S_{\text{Candidate}}=\{X_{\text{I}},X_{\text{II}},X_{\text{III}},X_{\text{aVR}},X_{\text{aVL}},X_{\text{aVF}},X_{\text{V1}},X_{\text{V2}},X_{\text{V3}},X_{\text{V4}},X_{\text{V5}},X_{\text{V6}}\}$, $S_{\text{Target}}=\{\mathrm{\varnothing }\}$, $F{1}_{\text{best, 0}}=0$

For i in 1:12

For $X_{\text{ECG}}\hspace{0.17em}$ in $S_{\text{Candidate}}$

Calculate $F{1}_{\text{ECG}}$ by training a model using $S_{\text{Target}}+X_{\text{ECG}}$ for 10 times

Calculate $P_{\text{ECG}}$ by conducting t-tests between $F{1}_{\text{best},i-1}$ and $F{1}_{\text{ECG}}$

If min $(P_{\text{ECG}})$ < 0.05

Xbest,i = argmin(PECG)

Remove $X_{\text{best},i}$ from $S_{\text{Candidate}}$, Add $X_{\text{best},i}$ to $S_{\text{Target}}$

$F{1}_{\text{best},i}=F{1}_{X_{\text{best},i}}$

Else

Break

Output $S_{\text{Target}}$ as the optimal subset