Feature selection plays a crucial role in enhancing efficiency of machine learning models by reducing complexity and improving accuracy. While deep learning and filter methods exists, nature-inspired algorithms offer distinct advantages for feature selection. They can explore high-dimensional, non-linear search spaces, avoiding local optima, and adapt to diverse problem domains without using extensive labeled data or predefined assumptions. Binary Peacock Algorithm (bPA) is an established nature-inspired algorithm that mimics the lekking and mating behaviour of peacocks and peahens to balance exploration and exploitation. bPA has demonstrated its potential in feature selection tasks over other algorithms. Despite its ability to balance exploration and exploitation effectively, bPA often suffers from slow convergence and higher computation time, making it less efficient for large-scale or time-sensitive applications. Therefore, improvements in population movement can be done to boost its speed and make it more useful in real-world scenarios. This paper proposes three enhanced variants of bPA: (1) VbPA which integrates velocity dynamics to enhance population movement, (2) SbPA that incorporates spiral movement in the search solution space, and (3) CWbPA:- a combinational walk strategy to mobilise population in search space. Experimental results on benchmark datasets show that VbPA and CWbPA significantly improved computation time and accuracy, while SbPA performed the worst across most metrics. CWbPA demonstrated accuracy gains up to 14.3%, with feature reductions of up to 47.8% and over 50% reduction in computation time. These results show that including structured velocity dynamics in bPA improves its performance making it more efficient for binary problems such as feature selection.