Machine Learning Models for Predicting Hypothyroidism: Utilizing Synthetic Data for Improved Accuracy

Abstract

This study presents a novel approach to early hypothyroidism detection by integrating synthetic data generation with machine learning (ML) techniques. Facing the challenge of limited and imbalanced healthcare datasets, we employ synthetic data to augment training sets, ensuring a richer and more diverse data pool for ML application. Key indicators of early hypothyroidism are distilled through feature selection, optimizing ML model inputs. We test various ML classifiers, including Support Vector Machines (SVM), Random Forests (RF), and Gradient Boosting Machines (GBM), demonstrating enhanced diagnostic accuracy with our approach. Initial outcomes suggest that combining synthetic data with ML significantly boosts early detection capabilities, offering a promising direction for overcoming traditional data scarcity in medical diagnostics.