Document Classification in HEIs Using Deep Learning

A CNN, RNN, and Hybrid CNN-RNN Approach

Abstract

 Higher Education Institutions (HEIs) are increasingly confronted with the complexities of evolving rules and requirements, necessitating innovative technology solutions to streamline document handling processes. Traditional paperwork methods are often inefficient and error-prone, leading to potential non-compliance. This research addresses these challenges by developing an AI-powered electronic document management system designed to automate compliance checks and simplify document handling as HEIs grow. The primary objective is to create a document classification model utilizing deep learning techniques, including Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and a hybrid CNN-RNN approach, to enhance document accuracy and compliance. The study involves collecting and preprocessing a substantial dataset of documents, designing and evaluating various deep learning models, and optimizing hyperparameters. Performance comparisons among the models indicate that the hybrid CNN-RNN architecture outperforms individual models, achieving superior accuracy, recall, and F1-score, alongside a significantly lower mean squared error (MSE). Initial evaluations revealed the CNN, RNN, and CNN-RNN models achieved accuracies of 73%, 44%, and 27%, respectively, on the raw dataset. However, with an upgraded dataset, these models improved to 76%, 48%, and 79% accuracy, respectively, highlighting the hybrid model's enhanced capability in accurately classifying documents. The findings revealed the effectiveness of integrating advanced deep learning techniques to improve document verification processes in HEIs, ultimately facilitating better compliance and operational efficiency.