Binary classification of Lupus scientific articles applying deep ensemble model on text data

Lupus as a chronic autoimmune disease is a difficult disease to diagnose, where more than 60 percent of people with lupus were incorrectly diagnosed. Accurate and early diagnosis of lupus is a daunting challenge, where its signs and symptoms usually mimic some other diseases and may appear as either permanent or temporary signs. On the other hand, tons of scientific articles are published, accumulating valuable information, such as lupus prevention, diagnosis, and treatment plans. The availability of big data scientific articles plus deep learning text analytics methods makes it now possible to harness this data to advance Lupus research in general, providing timely fashion and high-quality information for Lupus diagnosis and treatment. In this work, we develop deep learning text categorization techniques on top of the PubMed articles to automatically classify Lupus scientific articles, demonstrating the potential for mining large-scale scientific articles with real-time update by new articles published in a daily basis. Using ensemble deep learning models help us improve the weakness of individual deep learning models when it comes to diagnostic classification. In the proposed deep ensemble model, Majority scheme is used for the output results of applied individual models, LSTM, CuDNNGRU, RNN and CNN, aiming to predict a class per each sample.