Diabetes monitoring and prediction using machine learning techniques

Abstract

Diabetes is a chronic disease characterized by an increase in blood sugar levels. Diabetes, if left untreated, causes devastating body complications such as heart attacks, nerve damage, blindness, kidney failure, and limb amputations among others, which may lead to death. Detecting and treating diabetes at an early stage is critical for lowering the risk of serious complications and keeping diabetics healthy. Various prediction algorithms were employed in this study to predict diabetes on a dataset containing 1000 rows and 8 features. We combined ensemble learning techniques such as Cat Boost Classifier and LGBM Classifier with K-Nearest Neighbor (KNN), Naive Bayes (GNB), Support vector machine (SVC), Logistic regression (LR), decision tree (DT), and Gaussian NB. Accuracy, recall, precision and f1 score were all used to evaluate each model. With an accuracy of 90%, the LGBM Classifier was the first model to perform well followed by Cat Boost Classifier (88.5%), SVC (86%), K Neighbors Classifier (85.5%), Decision Tree Classifier (83%), Logistic Regression (83%), and Gaussian NB (79%). Machine learning is helping to improve the health sector in a variety of ways, including disease prediction, which has helped to reduce death rates and complications. Using machine learning aids in identifying hidden information that traditional methods could not identify. Screening people is critical for identifying people who are asymptomatic but at risk of developing diabetes. As a result, machine learning (ML) techniques can be used on new registered patients' data sets to detect disease at an early stage, assisting physicians in their decision making.