Machine Learning-Based Prediction Model for Thyroid Cancer Diagnosis Using Clinicopathologic Features

Thyroid cancer is the most prevalent endocrine tumor, with an increasing global prevalence. Although the prognosis is generally positive, existing diagnostic tools are limited in their ability to detect conditions early and accurately, thus obscuring a suitable treatment. Consequently, a timely and accurate diagnosis is essential to improve patient outcomes. This study seeks to create a machine learning predictive model for the classification of thyroid cancer as benign or malignant utilizing clinicopathologic data. The data set comprises 13 variables, which include demographic information from the patient, medical history, genetic markers, lifestyle factors, clinical symptoms, and results of diagnostic tests. Three machine learning algorithms, Random Forest, Logistic Regression, and XGBoost, were executed and assessed utilizing the confusion matrix. The models were constructed with Python and the Scikit-learn framework. XGBoost attained the best accuracy at 82.49%, succeeded by Random Forest at 82.28% and Logistic Regression at 76.80%. The results illustrate the ability of machine learning models trained in clinical data to precisely predict thyroid cancer, facilitating early identification and improving clinical decision-making.