The images are converted to greyscale and have their pixel values normalised to be between 0 and 1. Using digital images or scanned documents, optical character recognition (OCR) software can automatically recognise and categorise handwritten and printed text. However, it can be difficult to recognise and categorise handwritten characters because they may vary in size, form, and style. In recent years, convolutional neural networks (CNNs) have been an effective method for achieving outstanding performance in OCR tasks. To classify and identify handwritten alphabets and digits, this research proposes a novel method that makes use of a CNN-based support vector machine (SVM) classifier. The method is tested using the EMNIST dataset, which consists of 240,000 images of handwritten upper and lowercase letters, numbers, and distinctive symbols. On the EMNIST dataset, the suggested method obtains an accuracy of 88.1%, which is a notable improvement over conventional OCR methods that depend on feature extraction and template matching. The CNNSVM model is more resistant to differences in handwriting style since it learns features directly from the images. The suggested CNN-SVM model also does away with the time-consuming, error-prone processes of feature extraction and template matching. Additionally, the CNN-SVM model is easily adaptable to identify and classify different kinds of characters, including symbols or alphabets that are not written in English.