<p>Feature selection plays a pivotal role in machine learning, aiming to reduce data dimensionality and enhance classification accuracy by identifying the most informative features. In this work, we propose a novel multi-strategy fusion approach, the Novel Binary Equalization Optimizer (NBEO), designed to address high-dimensional feature selection challenges. The NBEO extends the Binary Equalization Optimizer (BEO) by incorporating two key enhancements to improve optimization efficiency. First, reverse learning and Tent chaotic map operators are integrated to increase population diversity and improve the quality of the search solution. Second, we introduce six dynamic transfer function adjustment methods to broaden the solution search space. In our experiments, eight types of chaotic mappings and reverse learning mechanisms are embedded into the population initialization phase of the equalization optimizer (EO), forming nine multi-strategy fusion BEO variants. Additionally, we explore eight transfer functions from classical S-type, V-type, U-type, and Z-type families to construct various BEO variants. The most effective NBEO variant is further optimized with six dynamic adjustment methods. We conduct comparative studies against five swarm intelligence-based optimization algorithms and four top-performing algorithms from the CEC competition. Simulation experiments are performed on 20 high-dimensional standard UCI datasets and 12 high-dimensional cancer gene expression datasets, with statistical analysis of the results. Our findings demonstrate that the proposed method significantly expands the search space, reduces computation time, and improves classification accuracy across a majority of the datasets.</p>