Machine Learning for Diabetes Forecasting

Abstract

The medical industry and machine learning now have a stronger connection because to technological advancements. This work uses machine learning to forecast the prevalence of diabetes, a worldwide illness. Predicting the disease at its early stages is the goal in order to make treatment or management of the illness easier. I have worked with a dataset that has nine characteristics and 100,000 occurrences. This dataset includes data on blood glucose level, BMI, age, smoking history, heart disease, gender, and HbA1c level in addition to hypertension. These are the principal markers of diabetes. I have used the Random Forest Classifier to forecast the sickness. My study's findings have been contrasted with those of other machine learning techniques, such as Decision Tree Classifier and Logistic Regression. After comparison, it was discovered that, out of all of these methods, the Random Forest Classifier had the greatest accuracy and AUC score. Using this approach, I have determined that the accuracy is 95.67%, with an AUC score of 0.97. KNN has 88% accuracy, decision trees have 94%, and logistic regression has 88%. The findings demonstrate our study's remarkable success in accurately predicting diabetes. My research indicates that there is great potential for integrating computer science and medicine so that dangerous conditions like diabetes may be identified early on.