In a Feature Subset Selection (FSS) problem, the objective is to obtain an optimal feature subset on which the learning algorithm can focus and neglect the irrelevant features. A wrapper formulates the FSS as a combinatorial optimization problem. In this paper, we propose an elitist quantum inspired Differential Evolution (QDE) algorithm for FSS. The performance of QDE is found to be significantly better than that of Binary Differential Evolution (BDE) algorithm on three benchmark problems taken from literature. In both cases, logistic regression was chosen as the classifier. Further, QDE outperformed not only the extant algorithms reported in literature but also the t-statistic cum logistic regression based filter method for FSS.
https://link.springer.com/chapter/10.1007/978-3-319-26181-2_11