<p>Medical diagnosis datasets are often high dimensional and complex, requiring effective feature selection to reduce dimensionality while improving classification accuracy. This research presents the multi-objective improved binary differential evolution algorithm for FS (MOBDE-FS) that is designed for medical diagnosis datasets. MOBDE-FS includes several modifications to the original DE algorithm. First, it produces distributed, diverse initial candidate solutions to an optimization problem using the refraction learning technique. Second, it utilizes three mutation functions based on opposition-based learning, Gaussian perturbation, and Lévy flight, which are applied across generations to enhance solution diversity and facilitate the discovery of non-dominated solutions. The performance of the MOBDE-FS was examined on nine real-world UCI medical datasets and compared with seven popular, effective optimization algorithms. The simulation results showed that MOBDE-FS performed better than other algorithms in eleven popular evaluation measures: classification error rate, accuracy, number of selected features, Pareto front quality, execution time, hypervolume, generational distance, inverted generational distance, Pareto front deviation, Pareto front approximation error, and spread. Statistical analyses conducted through the Friedman and Wilcoxon signed-rank tests confirmed the consistent superiority of the MOBDE-FS.</p>