多种机器学习算法在甲状腺癌复发风险评估中的对比分析

Comparative Analysis of Multiple Machine Learning Algorithms for Risk Assessment of Thyroid Cancer Recurrence

摘要: 甲状腺癌术后复发风险的准确评估对改善患者预后及优化医疗资源配置具有重要意义。针对传统统计学方法在处理复杂非线性临床数据时的局限性，本文旨在探讨并比较多种机器学习算法在甲状腺癌复发预测中的应用价值。本文采用Borzooei和Tarokhian提供的临床数据集，包含17项特征变量，在对数据进行清洗及Label Encoding数值化处理后，按3:1比例划分为训练集与测试集，构建了逻辑回归、K近邻、决策树、支持向量机(SVM)以及随机森林、XGBoost、CatBoost共七种机器学习模型。通过受试者工作特征曲线(ROC)、曲线下面积(AUC)以及准确度、灵敏度、特异度等这一多维指标体系，全面评估各模型的分类性能。实验结果显示，七种模型均表现出优异的预测性能，AUC值均超过0.93，准确度均高于0.89。其中，集成学习算法表现最为突出：随机森林(Random Forest)以0.9904的最高AUC值展现了最优的泛化能力；XGBoost与CatBoost在整体准确度上并列第一(0.9375)，且XGBoost在特异度(0.9000)上表现最佳。特征分析进一步揭示，风险等级(Risk)、治疗反应(Response)及TNM分期是影响复发预测的核心指标。机器学习技术，特别是以随机森林和XGBoost为代表的集成学习算法，能有效提升甲状腺癌复发风险预测的准确性，该模型可作为一种客观、高效的辅助诊断工具，为临床医生制定个性化随访策略提供科学依据。

Abstract: Accurate assessment of postoperative recurrence risk in thyroid cancer is of great significance for improving patient prognosis and optimizing medical resource allocation. Addressing the limitations of traditional statistical methods in handling complex nonlinear clinical data, this paper aims to explore and compare the application value of various machine learning algorithms in predicting thyroid cancer recurrence. This study utilizes the clinical dataset provided by Borzooei and Tarokhian, comprising 17 feature variables. After data cleaning and numerical processing via Label Encoding, the data was split into training and testing sets at a 3:1 ratio. Seven machine learning models were constructed, including Logistic Regression, K-Nearest Neighbors (KNN), Decision Tree, Support Vector Machine (SVM), Random Forest, XGBoost, and CatBoost. The classification performance of each model was comprehensively evaluated using a multidimensional metric system involving Receiver Operating Characteristic (ROC) curve, Area Under the Curve (AUC), accuracy, sensitivity, and specificity. Experimental results demonstrated that all seven models exhibited excellent predictive performance, with AUC values exceeding 0.93 and accuracy surpassing 0.89. Notably, ensemble learning algorithms performed the best: Random Forest demonstrated optimal generalization ability with the highest AUC of 0.9904; XGBoost and CatBoost tied for the highest overall accuracy (0.9375), with XGBoost achieving the best specificity (0.9000). Feature analysis further revealed that Risk level, Response to therapy, and TNM stage were the core predictors affecting recurrence. In conclusion, machine learning techniques, particularly ensemble learning algorithms represented by Random Forest and XGBoost, can effectively improve the accuracy of thyroid cancer recurrence risk prediction. These models can serve as objective and efficient auxiliary diagnostic tools, providing a scientific basis for clinicians to formulate personalized follow-up strategies.