Algorithm 2. BgPCA-NMI

Start {Y: the projected data matrix via BgPCA} Instantiate the feature subspace to null, S0 = {Φ} Choose the first feature, Yj using Equation (6) via utilizing Equation (8) and execute $S_1=S_0 {\displaystyle \cup } Y_j$ For choosing the rest of the features, do Apply Equation (9) and update S Take S as the subspace of the top effective features