摘要
糖尿病是一种以高血糖为特征的慢性疾病,长期高血糖会导致机体代谢紊乱,进而损害心脏、肾脏等多种器官,严重影响患者健康。因此,糖尿病的早期预防极为关键。文章以Kaggle平台公开的糖尿病数据集为研究对象,首先运用Lasso方法进行特征选择,获取用于糖尿病预测的最优特征子集。然后,借助集成学习中的随机森林、AdaBoost和CatBoost模型进行糖尿病风险预测,结果显示CatBoost模型预测性能最佳。随后,通过5折交叉验证评估CatBoost模型的泛化能力,结果表明该模型在不同数据划分下具备稳定的预测性能。最后,利用SHAP方法对CatBoost模型进行解释性分析,发现空腹血糖是影响糖尿病风险的最重要特征,为糖尿病的早期预防提供了有效技术支撑。
Diabetes is a chronic disease characterized by high blood sugar.Long-term hyperglycemia can lead to metabolic disorders in the body,which in turn damages multiple organs such as the heart and kidneys,seriously affecting the health of patients.Therefore,early prevention of diabetes is of great significance.In the study,the publicly available diabetes dataset on the Kaggle platform is used as the research object.Firstly,the Lasso method is applied for feature selection to obtain the optimal feature subset for diabetes prediction.Subsequently,the random forest,AdaBoost,and CatBoost models in ensemble learning are used to predict the risk of diabetes.The results show that the CatBoost model has the best prediction performance.Then,the generalization ability of the CatBoost model is evaluated through 5-fold cross-validation.The results indicat that the model has stable prediction performance under different data partitions.Finally,the SHAP method is used to conduct an interpretability analysis of the CatBoost model.It is found that fasting blood glucose is the most important feature affecting the risk of diabetes,providing effective technical support for the early prevention of diabetes.
作者
陈路遥
胡坤
秦萌
孙乾
王希胤
CHEN Luyao;HU Kun;QIN Meng;SUN Qian;WANG Xiyin(North China University of Science and Technology,Tangshan Hebei 063210,China)
出处
《信息与电脑》
2025年第8期52-54,共3页
Information & Computer
基金
大学生创新训练课题“基于集成学习的钢铁工人患糖尿病风险可解释预测模型”(项目编号:X2024116)。
作者简介
陈路遥,女,本科。研究方向:疾病风险预测。
