Lifetime Prediction Of Breast Cancer Patients Based On Survival Stacking

Breast cancer is one of the most common malignancies in women and is a highly heterogeneous disease.It is often difficult to diagnose and treat based on clinical features alone.Early prediction of breast cancer,which has a high risk of long-term death,is of great significance for treatment and prognosis.It is difficult to analysis the high-dimensional and censoring data in classical regression framework.To address this dilemma,this dissertation devotes the following topics:(1)Selection of the significance clinical variables based-on Kolmogorov-Smirnov test;(2)Variable selection for high-dimensional genetic data upon Random Survival Forest importance and Pearson Correlation;(3)Prediction of the survival probability curve based on survival stacking technology by classification method.The categorical clinical features are selected upon Kolmogorov-Smirnov comparison,and the combination of importance index from Random Survival Forest and Pearson Correlation between features is explored to select the numerical features.The data set after pre-processing is transformed into categorical data using survival stacking technology.Machine learning methods such as Naive Bayes,Decision Tree,K-Nearest Neighbor,Adaptive Boosting,and Gradient Boosting Decision Tree are used to predict the survival probability curves.They are also compared with traditional survival analysis methods such as Cox proportional hazard model,Nnet-Survival,and Deep Hit.The consistency index modeled by survival stacking technique and classification methods on this data set are better than that of the state-of-art survival analysis methods,and the best one is K-Nearest Neighbor classification with consistency index of 0.847.