Machine Learning Models for Predicting Thyroid Cancer Recurrence: A Comparative Analysis

Abstract

Thyroid cancer is considered the most common malignancy of the endocrine system and encompasses a broad category of diseases that involve abnormal growth in thyroid cells. Thyroid carcinoma accounts for about 3% of the total cases of cancer diagnosis in the USA. The chief objective of the research project is to design and compare the performance of machine learning models in the prediction of thyroid cancer recurrence to overcome the limitations observed in the current predictive tools. This study aims to develop and compare Machine Learning models. In particular, this study considered different machine-learning algorithms to identify which model can effectively forecast the recurrence of thyroid cancer.  The dataset used for the analysis was from Kaggle, the ‘Thyroid Gland Dataset.’ This source had a very elaborate dataset, containing records of patients who were diagnosed with thyroid issues, including demographic data on variables that would be needed to see the recurrence of any disease. Besides, it contained demographic information about the patients, which would serve to comprehend population trends in the patients; examples are age, gender, and ethnicity. The clinical history data included size, histological subtype, lymph node involvement, and staging at diagnosis. This comparative analysis mounted a variety of machine learning algorithms, each of which was chosen based on its capabilities to face structured medical datasets for robust predictions. Each model was chosen based on their different strengths that correspond to characteristics in the dataset and the general goals of the prediction problem. Performance metrics used for the models included overall accuracy, precision, recall, and the F1 score. Logistic Regression performed slightly better than the random forest and the support vector machines. However, this difference in accuracy was minimal and all three can make quite accurate predictions on this data.  Logistic Regression provides transparency and interpretability, Random Forest provides high versatility and robustness, while SVM offers precision for complex relationship modeling. The integration of machine learning predictive models into clinical practice has great potential to transform decision-making, particularly in the management of thyroid cancer and the risk of recurrence. These models will greatly assist clinicians by consequently advising them on which patients have a high chance of recurrence, so early intervention might be considered and follow-up care given as need sets in.