Convolutional Neural Network Using Regularized Conditional Entropy Loss (CNNRCoE) for MNIST Handwritten Digits Classification

Convolutional Neural Networks (CNNs) have demonstrated remarkable performance in various image classification tasks, including the classification of handwritten digits in the MNIST dataset. It is evident that it has a vast go, but it too faces certain limitations such as overconfidence, complexity and overfitting. Hence, an approach leveraging Regularized Conditional Entropy (RCE) as loss function within a CNN architecture for enhanced accuracy in digit classification as a whole called Convolutional Neural Network using Regularized Conditional Entropy Loss (CNNRCoE). The regularization technique employed aims to mitigate overfitting and improve generalization by penalizing the complexity of the model. In this study, extensive experiments have been conducted on the MNIST Handwritten digits inclusive of 5 datasets to evaluate the effectiveness of the proposed method, comparing it with traditional CNN architectures. The results demonstrate that the integration of RCoE within the CNN framework yields superior performance, achieving state-of-the-art accuracy at about 98% with the increase about 20%, while maintaining sturdiness against overfitting.