Performance Analysis of a Handwritten Digit Recognition Using CNN-WOA

Handwritten Digit Recognition (HDR) techniques are acquiring favor in academia and business. Because of their complex and irregular patterns, handwritten texts are difficult to recognize. Word retrieval in limited-resource scripts is complex and time-consuming. Randomizing samples and increasing the size of training datasets enhance deep learning and neural network, models. HDR techniques heavily rely on manually-programmed key point information. The writing styles and dimensions of the input sample make numerical identification and categorization difficult. This work circumvents these limitations by employing a CNN model and an effective HDR scheme. HDR enhances the classification accuracy of the MNIST dataset. The Gaussian filter and the Convolution algorithm polish the numbers. CNN-WOA is indicated as the model because the Algorithm for Whale Optimization Algorithm (WOA) fine-tunes CNN parameters. Using accuracy, precision, recall, specificity, AUC, F1-score, and false positive rate, simulation outcomes of the CNN-WOA model are compared to state-of-the-art techniques.