Determining Treatment Dosage for Hypothyroidism Using Machine Learning

Abstract

Hypothyroidism is a prevalent chronic condition requiring precise levothyroxine dosing to prevent complications. However, factors such as stress and weight fluctuations complicate dosage determination. This study applies machine learning to improve dosage prediction accuracy. A synthetically generated dataset incorporating key clinical parameters (age, gender, TSH, T3, and T4) was used to train and evaluate predictive models. Compared to the current standard-Poisson Regression (64.8% accuracy), our approach achieved significant improvements: Ridge and Lasso Regression (82%), Support Vector Regression (83%), and k-Nearest Neighbors (86%). These results highlight the potential of machine learning in optimizing hypothyroidism treatment and enhancing patient outcomes.