Bengali Query Processing System for Disease Detection using LSTM and GRU

Abstract

This paper proposes a disease detection system where it receives the query in form of symptoms of the disease in the Bengali language. This system is able to handle natural language queries in Bengali. The proposed system assists a layman to detect a probable disorder or disease in their body using disease symptoms. The proposed research work is challenging due to insufficient resources in vernacular languages like Bengali. This system receives a description of the patient's symptoms in the Bengali language and after processing the natural language text, it detects any potential disorders or diseases that may have occurred. This research work has been implemented separately by using the two most popular sequential prediction models. One is Bi-directional Long Short-Term Memory (Bi-LSTM) and the other is Bi-directional Gated Recurrent Unit (Bi-GRU). Both Bi-GRU and Bi-LSTM have provided significant results on a dataset of 3714 samples. The raw clinical text categorization data has been gathered from the Kaggle to build the detection model. The performances of disease detectability of both models have been measured using precision, recall and f1-score. The accuracy of the proposed system using the Bi-LSTM and the Bi-GRU models are 97.85% and 99.73%, respectively.