ALT & AST prediction using optimized GAN model based on diabetes and metabolic function data of type 2 diabetes mellitus person

Abstract

 The incidence of Type 2 Diabetes Mellitus causes metabolic dysfunction and severe health complications. In this paper, optimized machine learning approach predicts liver enzyme levels and enhance diabetes management. The proposed pre-processing algorithms are (i) semi graph theory (SGT) based relation data extraction (ii) Generalized Rough set theory (GRST) for redundant data reduction. The proposed liver enzyme prediction algorithm is (i) Bayesian optimised Sinusoidal regression (BO-SR) (ii) Genetic Algorithm optimised Polynomial Regression (GA-PR). The proposed method applied in the clinical data such as demographics, metabolic function, and sleep data. Semi-graph theory is used for relational structuring and generalized rough set theory is used for eliminating redundant information. Adam optimised Generative Adversarial Networks are used to generate synthetic Thermic Effect of Food data and improve data size and diversity. Genetic Algorithm and Bayesian Optimization ensure optimal parameter selection and improves prediction accuracy. The proposed method has an accuracy of 97.3% in predicting Aspartate Aminotransferase (ALT) and Alanine Aminotransferase (AST) levels, this ensures early diagnosis and intervention for diabetes-related liver complications.