Improved leaps and bounds variable selection algorithm based on principal component analysis

• We put forward an improved leaps and bounds regression algorithm.
• The improved algorithm can be applied when the descriptors are more than samples.
• Optimal subset can be obtained in the algorithm in a significantly short time.

In this paper, a new variable selection algorithm is described, based on leaps and bounds regression. The algorithm removes the limit of the traditional algorithm that the descriptors must be less than the samples, by replacing the original variables in a subset evaluation with a small number of principal components. Two different sizes of variables data sets were employed to investigate the performance of the new algorithm. The result shows that the improved algorithm can obtain optimal or good sub-optimal subsets when a different number of principal components are used.