Deep Learning Architecture Based Document Classification System

Document classification system has the ability to instinctively categorize documents into one of the established classifications and is important for indexing and retrieval applications in office environment and educational scenarios. Several profound learning and machine learning strategies had been discussed in literature to the categorization of documents. Many automated approaches for classification of documents in English languages are widely reported in literature. However development of methods for classification of documents in regional languages like Tamil is still in its infancy. It is an approach based on design to stay deep machine learning is proposed for the classification of documents in Tamil language. The approach involves creation of document dataset and document classification model followed by evaluation of the performance of developed model. The document dataset is created by collecting and digitizing documents from five different categories namely book, circular, invitation, newspaper, letter. 2D CNN and VGG16 architectures are employed in instruction the document categorizing the model. Then the evaluated performance of both models using measures of precision, recall, F1 score, and precision. The results of the comparison, which VGG16 model perform with higher accuracy of 98.8% in classifying digitized Tamil documents.