Hybrid Machine Learning Framework for Type 2 Diabetes Prediction Using Metaheuristic Optimization Algorithms

Abstract

The general basis of diabetes prediction using machine learning involves the application of algorithms that take an overall look at multiple features like BMI and glucose levels, age, genetic predispositions, and other conditions that may predict the likelihood of developing diabetes. The data-driven schemes, such as neural networks or DTs, find patterns in past data and use these to provide reliable predictions about future diabetes cases. These schemes keep learning and improving; they grow with new inputs. ML now helps in early detection by the use of large datasets, thus enabling early actions such as lifestyle changes or medical therapies. Finally, it enhances healthcare by providing individualized risk assessment and thus enables timely actions to diminish the burden of diabetes. In addition, the application of ML schemes, including Gaussian Process Classification-GPC, Linear Discriminant Analysis-LDA with Henry Gas Solubility Optimization-HGSO, Chaos Game Optimization-CGO, and Chef-Based Enhancement scheme-CBOA, has greatly benefited the process of prediction. These schemes were combined with optimizers, guided by the objective of this work, which deals with predicting the type of diabetes and the diagnosis of persons vulnerable to it. This was a strategic fusion aimed at creating new hybrid schemes with increased precision in prediction. Further analysis showed that the GPCB model was the best, with an impressive 0.981 during training. By contrast, the GPCG and GPHG schemes are relatively less accurate, with an accuracy of 0.963 and 0.946, respectively. These results justify the utility of the integrated approach, where advanced ML algorithms were able to generate predictive schemes superior in terms of accuracy and efficiency compared to the classical methods.