Abstract
Hypothyroidism, a prevalent chronic health condition, can lead to
serious complications if untreated. Management typically involves
synthetic thyroid hormone replacement, with dosage being crucial for
effective treatment. However, factors like stress and weight
fluctuations impact thyroid hormone levels, posing challenges in dosage
determination. This study introduces an innovative approach using
machine learning for precise dosage prediction. We developed a synthetic
thyroid disease dataset, encompassing parameters such as age, gender,
TSH, T3, and T4, to train and evaluate various machine learning models.
The study aimed to surpass the current state-of-the-art in dosage
prediction, which is Poisson Regression with a 64.8% accuracy. Our
findings reveal that Ridge Regression and Lasso Regression achieved an
accuracy of 82%, while Support Vector Regression Machines attained
83%. Notably, k-Nearest Neighbour (k-NN) algorithm demonstrated the
highest accuracy of 86%, marking a significant improvement of over 21%
from the existing standard. This enhancement in prediction accuracy
holds potential for optimizing treatment efficacy and patient outcomes
in hypothyroidism management.