Digit recognition from handwritten data in the form of photographs has drawn increased interest in the area of recognizing the pattern because of its many applications and inconsistent learning methods. A classification algorithm states that recognizing the digits and extracting characteristics are the two main processes needed for handwritten digit identification. The proposed project was designed to improve handwritten digit recognition accuracy and calculation speed, paving the path toward digitalization. In the current study, a convolutional neural network was used as a classifier, together with the DL4J framework, a dataset from MNIST that had adequate training and testing data, to identify handwritten digits. The approach indicated above successfully transmits accuracy up to 99.21% as compared to earlier suggested strategies. The method is improved by the proposed system, which also drastically reduces the computing time required for training and testing.