Bengali language-based Disease diagnosis system for rural people using Bidirectional Encoder Representations from Transformers

Abstract

Machine intelligence is an essential branch of computer science that imitates human behavior. A human recognizes every real-world object and incident by their distinguishable features. If a computerized system incorporates human cognitive abilities, the system will be able to classify many real-world phenomena accurately. Therefore, an automated system that detects the disease based on symptoms has been addressed. Bidirectional Encoder Representations from Transformers (BERT) has been used to design this system. The BERT model has been developed based on the Transformers concept, forecasting a token by looking at every token in the series. The proposed system recognizes each and every individual’s illness in the Bengali language. Punctuation marks are stripped from the user-supplied query which is in form of disease symptoms. Tokens are obtained from the individual’s query. Tokens that have been filtered are sent to the proposed system. The proposed method employs filtered tokens to diagnose the disorder. The BERT model is trained using a dataset of diseases and their symptoms obtained from Kaggle. Precision, recall, fl-score, root mean squared error (RMSE), mean squared error (MSE), and mean absolute error (MAE) have been used to describe the proposed system’s performance. The accuracy of the proposed system is 98.12% whereas RMSE, MSE, and MAE exhibit 1.8618, 1.0852, and 1.0826 respectively.