EFFICIENT DEEP CNN FOR HANDWRITTEN DIGIT RECOGNITION: A PYTORCH AND SKORCH APPROACH

Handwritten digit recognition remains a benchmark problem for evaluating machine learning and deep learning algorithms. In this study, we propose a simple yet effective convolutional neural network (CNN) architecture for multi-class classification of the MNIST dataset, implemented with PyTorch and Skorch. The model employs hyperparameter optimization via grid search to fine-tune the activation functions, kernel sizes, optimizers, and dropout rates. The experimental results demonstrate state-of-the-art performance, achieving 99% accuracy, precision, recall, and F1-score across all 10 digit classes. Comparative analysis against baseline models—including decision trees, SVMs, ANNs, and conventional CNNs—confirms that the proposed model consistently outperforms prior approaches. These findings highlight the effectiveness of lightweight, well-tuned CNN architectures for digit recognition tasks and demonstrate the utility of PyTorch and Skorch as efficient frameworks for model design, training, and deployment