In this research, a modification of the Harris hawk optimization (HHO) is introduced and applied to feature selection. The proposed HHO variant is termed as hybrid multi-leader HHO with differential evolution (MLHHDE). To help hawks overcome the situation of falling into local optimum during the searching process, an improved memory structure is introduced, which makes the hawks learn simultaneously from the historical best and global best positions. Besides, the multi-leader mechanism is also introduced to make full use of the valuable information from global best memories (leaders), enhance the diversity of hawks’ search mode, and improve the hawks’ exploration capability. Furthermore, differential evolution operations are integrated into HHO to further improve the weak exploration phase. Our proposed MLHHDE algorithm integrates a V-shaped transfer function, which can convert continuous solutions to binary solutions. To validate the modification of MLHHDE, we compared its performance with the advanced optimization algorithms through three experiments. In the first experiment, the performance of MLHHDE to solve a set of problems from the CEC 2017 benchmark is evaluated. Meanwhile, the second experiment aims to apply the binary version of MLHHDE to tackle the feature selection task by applying it to a set of sixteen datasets from the UCI repository. In the third, we applied the proposed model as a quantitative structure-activity relationship method to predict the influenza viruses H1N1 as a real-world application. The performance of the proposed MLHHDE is assessed using a number of evaluation indicators. The experiment results prove the powerful capability of MLHHDE to find the optimal solution in the two experiments as well as it outperforms other methods (i.e., either global optimization or feature selection). In addition, the developed MLHHDE provides accuracy overall the UCI datasets nearly 84% with difference 5% between it and traditional HHO, also, it provides accuracy 92% with standard deviation when applied to predict H1N1.