A Study on Predicting Postoperative Recurrence Risk of Differentiated Thyroid Cancer Based on Machine Learning

This study predicts the risk of postoperative recurrence of differentiated thyroid cancer based on a machine learning approach. The data were obtained from the clinical dataset of postoperative recurrence of differentiated thyroid cancer in the UCI Machine Learning Repository, which contains a total of 383 cases involving 17 clinical and pathological feature variables. The study was conducted through data preprocessing, including data cleaning, format normalization, feature reconstruction, and coding transformation of category-based variables, and finally the data were divided into training and testing sets. The study establishes three models, and processes the data by using standardization and normalization methods. The hyperparameters are optimized by grid search and the model performance is evaluated by accuracy, recall and AUC. The results show that the Multilayer Perceptron (MLP) model outperforms the other two models in preprocessing, standardization and normalization of data, with the highest accuracy of $\mathbf{0. 9 7 3 9}$ and the highest AUC value of $\mathbf{0. 9 9 5}$. This study demonstrated the effectiveness of machine learning methods in the prediction of recurrence risk after thyroid cancer surgery, and the MLP model showed good prediction performance, which provided strong support for clinical decision-making.