In high dimensional data, the Feature Selection (FS) approach plays an important role in overcoming accuracy, time complexity, and space complexity. This paper proposes a binary version of the hybrid two-phase multi-objective FS approach, based on Particle Swarm Optimization (PSO) and Gray Wolf Optimization (GWO). The first objective is to minimize the classification error rate, and the second objective is to reduce the number of selected features. In the first phase, the proposed approach performs the global search, and in the second phase, it goes for the local search. In the case of global search, i.e., for the exploration phase, the property of PSO is used. In the case of local search, i.e., for the exploitation phase, the proposed method uses the modified version of PSO and GWO. The particles are learning not only from the positions but also from the mass and acceleration, based upon Newton’s second law of motion. In some cases, while in the iterations for updation of particles, a stall arises. Due to which the particles are not updating their positions. To avoid the stall in the iterations, this paper introduces a new term, i.e., population factor, which can omit the stall in the iterations if it arises. The prominent features which are selected from the proposed approach are tested on five well-known classification algorithms. The performance of the proposed approach is investigated on eight high-dimensional gene expression data. Experimental results comparison shows that the proposed approach for FS performs efficiently and effectively than other metaheuristics, statistical, and multi-objective FS methods.