Algorithm 1: SP-LIME algorithm.

Inputs: $S$: The set of samples $F$: The set of features $X$: The set of instances for explanations Run the explanation model on all instances $x_i\in X$ with the aid of the LIME algorithm. Construct the explanation matrix $W_{ij}$ that represents the local importance of the interpretable features for each instance. Compute the global importance of each feature $f_j\in F$ with $I_j={\displaystyle \sum }_{i=1}^n\left|W_{ij}\right|$ Maximize the coverage function by iteratively adding the instance with the highest maximum coverage gain. Outputs: Feature importance $I_j$ Instances that cover the important features $V\subseteq S$