<p>Feature selection is a critical step in handling large-scale datasets, where the goal is to identify a minimal subset of features that preserves or enhances predictive performance. Traditional multi-objective evolutionary algorithms (MOEAs) often struggle with scalability, diversity preservation, and convergence speed when applied to high-dimensional data. To address these issues this work proposes a binary multi-objective neural network algorithm that integrates new learning strategies for large-scale feature selection problems. A novel fuzzy membership function is employed for effective binary encoding, while a novel biasing strategy enhances binary pattern learning. The binary form is employed on standard benchmark functions and compared with 5 state-of-the-art algorithms to validate its performance. In addition, a non-dominated sorting mechanism tailored for feature selection and a reference point strategy for diversity preservation are incorporated. The algorithm was evaluated on fifteen real-world datasets from the UCI repository, ranging from 60 to over 2,000 features. Experimental results show that the proposed approach improves classification accuracy and also reduces the average number of selected features. It also achieves faster convergence compared to NSGA-II, SPEA2, and other state-of-the-art MOEAs. These findings demonstrate the strong potential of the proposed method for tackling complex, large-scale feature selection tasks.</p>