Position-Based Feature Selection for Body Sensors regarding Daily Living Activity Recognition

Algorithm 1

for i = 1 : 3 % We run three feature selection algorithms

   CVO = LeaveOneOutCV ( ); %Leave-one-out cross-validation

   for j = 1:CVO. NumofTestSets

      % Get training and test sets

      X_train = Features {i} (training Index,:);

      Y_train = Activities (training Index);

      X_test = Features {i} (testIndex,:);

      Y_test = Activities (testIndex);

      k_idex = 0;

      for k = 1:CVO. MaxNumOfFeature (i)

         k_idx = k_idex + 1;

         % Get features according to rank of feature

         fs_index = FeatureRanking {i} (1 : k);

         % Evaluate the feature importance by K-NN algorithm.

         Prediction = knnModel (X_train (:, fs_index), ….

              Y_train, X_test (:, fs_index), 1);

         cvAccuracy (j, k_idx) = sum (double (Prediction == Y_test)) ….

                           /TestSize (j);

      end

  end

Accuracy {1,i} (1: size (cvAccuracy, 2)) = mean (cvAccuracy);

end