DiabChatbot: A Machine Learning Chatbot for Early Diagnosis of Type II/Mellitus Diabetes and Diet Recommendation in an Indian Scenario

Authors:

Abstract
It is predicted that by 2025, there will be 79.4 million Type 2 diabetics in India, up from the estimated 62 million currently living with the condition. Therefore, diabetes and its related health issues must be prevented through before-time detection and treatment. The goal of the research is to generate a chatbot specifically designed for the Indian context, which would offer the Indian population appropriate food recommendations and an automated early diabetes diagnosis. The ChatBot was created using the Diabetes Dataset for diabetes prediction. We've also put forth a DiabChatBot architecture for the success of a diabetes prediction chatbot powered by machine learning. The overall System flow of the DiabChatBot is also discussed here. The machine learning results for the PIMA dataset are compared with the dataset of Diabetes to showcase the effectiveness of the Diabetes dataset. On the dataset, several ML supervised and ensembling classifiers are used. Parameters such as Area Under Curve (AUC), Recall, Cohen Kappa Score, F-1 Measure, Accuracy, Sensitivity, Specificity and Precision are used to gauge the performance of algorithms for both datasets. With an AUC value of 0.975 and an accuracy of 0.983, the comparative findings unequivocally demonstrate that the CatBoost Algorithm performs exceptionally well in the Diabetes Dataset. Here the ROC graphs for both datasets are demonstrated as well. In the accuracy outcomes, the DiabChatBot, DiaBot and Health Care ChatBot are also evaluated for the Indian Scenario. The comparison unequivocally demonstrates that the DiabChatBot outperforms DiaBot and Health Care Chatbot in all ML techniques with an accuracy of 98%.

This publication has 12 references indexed in Scilit: