Machine Learning Based Approach for Predicting Diabetes in Young People of Bangladesh

Abstract

Diabetes is one of the deadly chronic diseases that occurs when the sugar level in the blood increases abnormally due to the absence of insulin hormone. Untreated Diabetes could lead a human to his/her death. In Bangladesh, the threat of Diabetes is really a matter of concern and people of all ages and gender are suffering equally. My research focused on young people who are under the age of 36 in Bangladesh. By using Machine Learning I have built a model which can predict the possibility of having or not having Diabetes. The model that I have built was trained by previous data of diabetic and non-diabetic patients. These data were collected from Bangladesh. On experiment, these data were processed and analyzed by various data pre-processing techniques. Then some classic Machine Learning algorithms like Logistic Regression, Random Forest, K-Nearest Neighbors and Naive Bayes were used for building the model and the performance of each of them was measured using metrics like Prediction Accuracy on the testing and training data, Confusion Matrix, Sensitivity, Precision, F1 score, Recall, Specificity, ROC and AUC. Overall Random Forest performed better than others. So, the Random Forest model was chosen for the prognosis of the disease and to demonstrate the use of the model. For that, I have built a web and an android application that will take required input data from a user according to the model and predict whether the user has Diabetes or not.