Abstract
This study pioneers a sophisticated approach to handwritten digit recognition by developing a stacked ensemble model that integrates advanced CNN architectures, specifically VGG16, VGG19, and ResNet. This model harnesses the collective strengths of these diverse networks to optimize feature extraction processes, significantly enhancing the accuracy and robustness of digit recognition. Unlike conventional methods that employ single-algorithm solutions, our approach utilizes a stacked architecture to distill and leverage unique insights from each algorithm, effectively navigating the complexities of variable handwriting styles. Our ensemble model has been rigorously evaluated on the MNIST dataset, where it demonstrates a superior performance, achieving an unprecedented accuracy rate of 99%. This remarkable achievement not only validates the effectiveness of combining VGG16, VGG19, and ResNet within a stacked framework for feature extraction but also highlights the model’s potential to revolutionize practices in broader pattern recognition contexts.
DOI: 10.61416/ceai.v27i4.9454