Differential Evolution (DE) algorithm applications encounter exploration challenges at the boundaries of the convex hull of the population, hindering the discovery of optimal solutions. To address this obstacle, we propose a novel modification of DE that explicitly utilizes “Boundary Individuals” to ensure a thorough exploration of the boundary regions. We applied the improved algorithm, DEBI, to the problem of optimizing neural networks’ hyperparameters. Our results, with three healthcare datasets, show that this modification of DE resulted in superior performance, with more consistent and higher fitness scores across generations, as well as improved precision and recall. These findings support the hypothesis that boundary exploration can significantly benefit learning.