Type-I and Type-II Diabetes Disease Prediction by Handling Trade-Off Between Hyperplane Margin and Classification Error In Support Vector Machine

This study presents a systematic approach to diabetes prediction utilizing a dataset comprising essential patient parameters. The dataset includes details from 70 sets of data on diabetes patients, encompassing lifestyle, insulin, and glucose levels. Key features such as BMI, glycated hemoglobin, glucose levels, cholesterol ratios, and blood pressure were identified as pivotal indicators. Principal Component Analysis (PCA) was employed for feature selection, followed by the implementation of the Support Vector Machine (SVM) algorithm for classification. Optimal hyperparameters for the RBF kernel were determined as C=1 and γ=0.5, resulting in a remarkable accuracy of 97.622%. For the sigmoid kernel, C=2 and γ=0.4 led to an accuracy of 96.345%. This research highlights the significance of meticulous feature selection and hyperparameter tuning in developing precise diabetes prediction models, showcasing the substantial potential for enhanced healthcare decision-making.