Algorithm 4: AKDB

Input: Training set $\mathcal{T}$ with features $\mathcal{L}=\{X_1,\cdots ,X_n,C\}$ and k.  Output: AKDB model. 1 Calculate $I(X_i;C)(1\le i\le n)$ from $\mathcal{T}$ for each feature and AMI; 2 Calculate $I(X_i;X_j|C)(i\ne j)$ from $\mathcal{T}$ for every pair of features and ACMI; 3 Let $\mathcal{L}$ be a list which includes all $X_i$ in decreasing order of $I(X_i;C)$; 4 Initialize the network structure $\mathcal{G}$ = LearnStructure($\mathcal{T},\mathcal{L},k$);         // Algorithm 1 5 $\mathcal{G}$ = FeatureSelection($\mathcal{T},\mathcal{G},\mathcal{L}$, AMI);                      // Algorithm 2 6 $\mathcal{G}$ = DependenceSelection($\mathcal{T},\mathcal{G},$ ACMI);                    // Algorithm 3 7 return $\mathcal{G}$;