Unraveling the Complexity of Thyroid Cancer Prediction: A Comparative Examination of Imputation Methods and ML Algorithms

Abstract

Despite being relatively rare, thyroid cancer is being identified more often as a result of improved awareness and detection. Even if it has a high survival rate, it is crucial to comprehend its forms, risk factors, and therapies. Better results and prompt intervention are made possible by the early detection of thyroid cellular alterations made possible by evolving machine learning (ML) techniques. The USA Cancer Data Access System's Thyroid Cancer Factor Data, gathered from patient questionnaires, are used in this study. Missing values and imbalance in the dataset are addressed using resampling techniques (SMOTE, under-sampling) and imputation techniques (Median, KNN). To increase the accuracy of thyroid cancer prediction and improve early identification and prognoses for improved patient care, a comparative analysis of machine learning algorithms (ML) (Logistic Regression, LDA, KNN, Decision Tree, SVM, Naive Bayes) with imputation and resampling techniques is being conducted.