The "Curse of Dimensionality" induced by the rapid development of information science, might have a negative impact when dealing with big datasets. In this paper, we propose a variant of the sparrow search algorithm (SSA), called Tent L\'evy flying sparrow search algorithm (TFSSA), and use it to select the best subset of features in the packing pattern for classification purposes. SSA is a recently proposed algorithm that has not been systematically applied to feature selection problems. After verification by the CEC2020 benchmark function, TFSSA is used to select the best feature combination to maximize classification accuracy and minimize the number of selected features. The proposed TFSSA is compared with nine algorithms in the literature. Nine evaluation metrics are used to properly evaluate and compare the performance of these algorithms on twenty-one datasets from the UCI repository. Furthermore, the approach is applied to the coronavirus disease (COVID-19) dataset, yielding the best average classification accuracy and the average number of feature selections, respectively, of 93.47% and 2.1. Experimental results confirm the advantages of the proposed algorithm in improving classification accuracy and reducing the number of selected features compared to other wrapper-based algorithms.