COMPARATIVE ANALYSIS OF SUPPORT VECTOR MACHINE KERNELS FOR DIABETES RISK PREDICTION

Abstract

Diabetes mellitus is a critical global healthcare challenge requiring accessible, non-invasive early screening. This study presents a comparative evaluation of Support Vector Machine (SVM) algorithms using three kernels: Linear, Polynomial (degree 3), and Radial Basis Function (RBF). Using 30,000 patient records from the CDC BRFSS across nine clinical indicators, the data was preprocessed via Z-score standardization and an 80:20 split. Results show that while the Polynomial kernel achieved slightly higher accuracy (84.50%) and precision (80.55%), it suffered from high computational latency (53.95 seconds). In contrast, the RBF kernel demonstrated superior balance and generalizability, yielding 84.40% accuracy, 84.40% recall, and the highest overall F1-score of 78.23% with a highly efficient training latency of just 11.77 seconds. Additionally, mathematical simulations on a real patient test vector yielded a definitive posterior probability of 80.66%. Ultimately, these findings establish the RBF-SVM framework as an optimal, robust, and computationally efficient predictive engine for automated community-scale clinical decision support.