DIAGNOSIS OF THYROID CANCER WITH MACHINE LEARNING ALGORTIHMS

Abstract

In recent years, thyroid cancer has become the second most common type of cancer worldwide after breast cancer. Today, with the widespread use of fast computers and the development of Artificial Intelligence algorithms, early diagnosis and accurate treatment have improved the quality of life for patients and increased survival rates. Traditional methods for diagnosing thyroid cancer can have limited accuracy, leading to misdiagnosis. In medical diagnostics, Machine Learning, a subfield of Artificial Intelligence, is frequently used. This method is widely applied in the diagnosis of diseases. Machine Learning algorithms help analyze large amounts of data to provide accurate and precise diagnoses. In this study, a dataset related to thyroid disease was used to predict whether thyroid nodules are malignant or benign as a classification problem in Machine Learning, without the need for biopsies of the patients. The available dataset was trained using classification algorithms under Supervised Learning, and each algorithm was used to predict models with Stratified Train Test Split and Stratified K-Fold Cross Validation methods. These models were later used to make predictions on the current test dataset. The classification algorithms employed in the study included Linear Discriminant Analysis, Naive Bayes, K-Nearest Neighbors, Random Forest, Support Vector Machines, Decision Trees, Artificial Neural Networks, Logistic Regression, Adaboost, XGboost, Gradient Boost, LightGBM Boost, and Catboost, which were implemented using Python. When comparing the performance metrics obtained from the results, it was found that Naive Bayes algorithm achieved the highest accuracy value with 92.31%, while XGboost algorithm had the highest ROC/AUC value with 92%.