Machine Learning Algorithm-based Early Prediction of Diabetes

Abstract

The metabolic disorder diabetes sometimes referred to as diabetic mellitus affects the body's normal glucose levels. Researchers introduce and fully describe a new hybrid machine learning algorithm (MLA) for the prediction of diabetes. Outcomes are also compared with those of similar research. Early detection of diabetes is essential to take necessary precautions (e.g. altering eating habits, monitoring patient weight etc.), delay the onset of diabetes and somewhat reduce the death rate and facilitate healthcare professionals' decision-making in the prevention and management of diabetes. In order to identify the most useful features in the detection of diabetes, this research aims to develop a unique hybrid selection technique that combines heat maps and a correlation matrix. The proposed hybrid feature selection analysis was used to determine a diabetic data set containing 499 cases and 12 features. Support vector classification (SVC) performed better than other models with a 94% accuracy rate followed by Logistic Regression (LR), K-Nearest Neighbour (KNN), Random Forest (RF), AdaBoost and Decision Tree (DT), which are 83%, 89%, 90%, 75% and 68%, respectively. Our suggested proposed framework is more effective than any other MLA. Using sequential stratified K-fold cross validation, the performance of each MLA is assessed using the metrics of accuracy, F1 score, time taken and area under the receiver operating curve.