Table 1 Computing feature supports using Random KNN bidirectional voting

/* Return support for each feature */

p ← number of features in the data set;

m ← number of features for each KNN;

r ← number of KNN classifiers;

F _i ← feature list for i^thKNN classifier;

C ← build r KNNs using m feature for each;

Perform query from base data sets using each KNN;

Compare predicted values with observed values;

Calculate accuracy, acc, for each base KNN;

$F\leftarrow {\bigcup }_{i=1}^r{F}_i$; {F is the list of features that appeared in r KNN classifiers};

for each f ∈ F do

   C(f) ← list of KNN classifiers that used f;

   $support\left(f\right)\leftarrow \frac{1}{\left|C\left(f\right)\right|}{\sum }_{knn\in C\left(f\right)}acc\left(knn\right);$

end for