Finding Type 2 Diabetes Using Sparse Balanced Support Vector Machine in Electronic Health Records

Early Type 2 Diabetes (T2D) diagnosis is crucial for optimal integrated therapy and patient follow- up. There has been a huge growth in the quantity of accessible Electronic Health Record (EHR) data, and developments in Machine Learning (ML) methodologies have been as amazing. Maintaining and modelling a huge data set can cause over fitting, model interpretability, and computational expenditure. We used Sparse Balanced Support Vector Machine (SB-SVM) to detect type 2 diabetes in an original EHR dataset (named FIMMG dataset). In particular, we have chosen just those aspects of the EHR collected from a homogeneous group of participants before the diagnosis of T2D, including exemptions, examinations, and medicine prescriptions. The proposed solution is trustworthy compared to other ML and Deep Learning methods utilised by the state-of-the-art to handle this problem. The SB- SVM offers the best prediction performance- computation time tradeoff, according to the data. In addition, the induced sparsity makes the model more interpretable while automatically handling high-dimensional data and the normally uneven distribution of classes.