Enhancing Diabetes Prediction: An Improved Boosting Algorithm for Diabetes Prediction

Abstract

Diabetes is increasing gradually due to the inability to effectively use the human body's insulin, which threatens public health. People with diabetes who go undiagnosed at early stages or who have diabetes have a high risk of heart disease, kidney disease, eye problems, stroke, and nerve damage for which diabetes diagnosis is crucial to prevent. Our advanced machine learning algorithm is the gateway to a revolutionary possibility of detecting whether the human body has diabetes. Developed this method based on machine learning with one lakh data and the main objective of creating a new and novel diabetes prediction model named moderated Ada-Boost(AB) that can accurately diagnose diabetes. About 10 different classification methods are applied in this research such as Random forest classifier (RF), logistic regression (LR), decision tree classifier (DT), support vector machine (SVM), Bayesian Classifier (BC) or Naive Bayes Classifier (NB), Bagging Classifier (BG), Stacking Classifier (ST), Moderated Ada-Boost(AB) Classifier, K Neighbors Classifier (KN) and Artificial Neural Network (ANN). The crucial contribution is to find out the appropriate values for the different models using the hyper-parameter tuning process. We have proposed a new boosting model named Moderated Ada-Boost(AB) which is the combination of the hyper-parameter tuned random forest model and Ada-boost model. Different evaluation metrics such as accuracy, precision, recall, f1 score, and others are used to evaluate the performance of the models. Our proposed new boosting algorithm named Moderated Ada-Boost(AB) provides better accuracy than other models whose training accuracy is 99.95% and testing accuracy is 98.14%.