Unlocking the Potential of NLP-Based Machine Learning for Efficient Document Classification

Document classification is a domain that has seen a rise in demand during the last decade. With powerful deep-learning architectures such as state-of-the-art (SOTA) convolutional neural networks (CNN) models for document classification a swell as image classification, they are computationally intensive while also requiring large amounts of data. However, this is not always possible in real-life application scenarios. This paper aims to provide a new approach to classifying images of documents using natural language processing (NLP)-based feature extraction from document images using the optical character recognition (OCR) technique. Along with it, it provides an alternative machine learning approach that gives high accuracy-91 % accuracy in our best case given by the Support Vector Classifier-which is better than all the other machine learning and 2D-CNN models, as well as an extremely fast training time.