Algorithm 2BAM Algorithm:

Given a training dataset $\mathbb{X}$ with p features, a testing dataset $\mathbb{T},$ feature selection methods $\{F{S}_1,\dots ,F{S}_t\},$ a threshold parameter $k,$ and a learning algorithm $M.$ (i) For $i=1,\dots ,m,$ generate bootstrap samples, ${\mathbb{X}}_1,\dots ,{\mathbb{X}}_m$ of the training dataset $\mathbb{X}.$ (ii) For each feature selection method $F{S}_q\in \{F{S}_1,\dots ,F{S}_t\},$ get features score matrix ${\mathbb{L}}^{\left(q\right)}.$ (iii) Normalize the score matrices in (ii) as ${\overrightarrow{\mathbb{L}}}^{\left(q\right)},q=1,\dots ,t.$ (iv) Use the arithmetic mean scores to combine the matrices in (iii) into one score matrix $\overline{\overrightarrow{L}}.$ (v) Use the score matrix in (iv) to compute the aggregated scores $\{\overline{{\overrightarrow{\ell }}_{.1}},\dots ,\overline{{\overrightarrow{\ell }}_{.p}\}}.$ (vi) Based on the aggregated scores in (v), compute the corresponding rank vector $\mathit{r}=(r_1,\dots ,r_p).$ (vii) Based on the rank vector $\mathit{r},$ keep the top $100k\%$ of the variable set $\{V_1,\dots ,V_p\}.$ (viii) Based on the selected feature set in (vii), use the testing dataset $\mathbb{T}$ and a cross-validation technique to train and test the model $M.$