一种基于概率校正和集成学习的肠癌肝转移预测模型  

A PROBABILITY CALIBRATION AND ENSEMBLE LEARNING BASED COLORECTAL CANCER LIVER METASTASIS PREDICTION MODEL

作　　者：吕奕[1] 王清[1] 

机构地区：[1]复旦大学计算机科学与技术学院,上海200433

出　　处：《计算机应用与软件》2011年第9期48-51,共4页Computer Applications and Software

基　　金：国家重大基础研究计划(2005CB321905)

摘　　要：提出一种基于概率校正和集成学习的机器学习模型,用来预测患者肠癌肝转移的概率。首先将AdaBoost和Class-bal-anced SVM的概率结果进行校正,再将其结果和Logistic回归的预测结果进行集成,获得最终的预测结果。预测模型在复旦大学附属肿瘤医院的肠癌患者数据集上与其他算法如AdaBoost、Class-balanced SVM、Logistic回归算法进行了比较,结果显示该模型具有更好的AUC性能,更适合于医生的临床辅助诊断。模型的AUC性能在UCI数据集上进一步得到了验证。A probability calibration and ensemble learning based machine learning model is proposed to predict the probability of colorectal cancer liver metastasis.At first the AdaBoost and Class-balanced SVM probability results are calibrated;then those results are assembled with Logistic regression prediction results to produce the final prediction result.The prediction model is compared against other algorithms such as AdaBoost,Class-balanced SVM and Logistic regression on the colorectal cancer patient data set of Fudan University Shanghai Cancer Center.The result shows the proposed model bears better AUC performance and is more suitable for clinic aided diagnosis.The model＇s AUC performance is further verified on UCI data set.

关 键 词：生物信息学 医学信息学 机器学习 概率校正 集成学习 

分 类 号：TP391.77[自动化与计算机技术—计算机应用技术]