Prediction of Recurrence of Differentiated Thyroid Cancer with Hybrid SMOTE-Stacking Model

Differentiated thyroid cancer (DTC) is the most frequent form of thyroid cancer. Although this type of cancer shows a favourable prognosis, the risk of recurrence remains a critical concern. Early and accurate prediction of the risk of recurrence is essential to improve patient outcomes and minimize this risk. This study proposes a hybrid model combining SMOTE (Synthetic Minority Oversampling Technique) and a stacking ensemble approach to predict DTC relapse. First, the dataset is balanced using the SMOTE technique, ensuring equal representation across classes. Then, the overall accuracy of the model is improved by the stacking method, which combines the predictions of multiple classifiers. This model has been tested on a publicly available dataset, with impressive results such as an accuracy of 99.09% and an AUC-ROC score of 0.998.