An improved multi-island genetic algorithm (IMIGA) by integrating various common improvement methods, optimized through D-optimal and analysis of variance (ANOVA) techniques. IMIGA’s performance is validated against various state-of-the-art algorithms across CEC2020 test suite and classic engineering problems, affirming its superiority. IMIGA’s potential in industrial applications is demonstrated by employing it for optimal design in a micropositioning stage, leveraging machine learning (ML) for parameter optimization. With the optimal design parameters, driven by IMIGA and a surrogate ML model, the proposed stage attained a first natural frequency of 55 Hz and maximum strokes of 266.7 μm, 110.9 μm, and 0.7552 mrad in the x-, y-, and θ-directions, respectively. These experimental results closely align with the predicted outcomes from the design phase, highlighting the accuracy and reliability of the employed methodology. Such consistency between experimental measurements and predicted values underscores the efficacy of the design approach in achieving desired performance metrics.