Document classification using deep learning

Abstract

This study tackles the classification of business documents into six pre-defined classes, (invoices, receipts, delivery notes, purchase orders, quotations and others). Three machine learning models, in increasing complexity, are proposed, implemented and compared. The models comprise a term frequency based classifier, a TF-IDF based Multinomial Naive Bayes classifier and a TF-IDF based Artificial Neural Network classifier. The models are trained and tested using a synthetic business document dataset, which was created as part of this project. The Neural Network classifier obtained the highest overall classification accuracy, over 97.7%, and outperformed the other two models by at least 5% points. The results show how this task benefits from the implementation of deep learning.