Prediction And Features Analysis Of Breast Cancer Based On Deep Neural Network

Survival data is widely used in the field of biomedicine,and also has important applications in economics and other fields,and its biggest feature lies in its censored,which has extensive practical significance for the e?ective modeling of such data.For survival data,what needs to be predicted is whether the patient has the disease or the degree of disease progression of the patient,so as to make a corresponding diagnosis for the patient.In this thesis,the METABRIC dataset was used,considering the clinical characteristics and genetic characteristics of patients,and the feature screening was carried out by random survival forest algorithm,and the method of combining survival analysis and deep learning was used after data processing to establish a joint model of Cox model and Deep neural network.Taking the log-likelihood function in the Cox model as the objective function of the neural network,the Reduce LROn Plateau function and Early Stopping method are used in the neural network to reduce the learning rate,thereby improving the model.Comparing the accuracy of the obtained joint model with some classical machine learning models,the results have been significantly improved.This thesis also considers the black-box problem in machine learning algorithms,Shapley Additive Explanation Values can be used to perform explanatory analysis of model features,and can clearly explain the positive and negative e?ects of various variables on patients’ conditions and the overall prediction trend according to the resulting figures,which further demonstrates the e?ectiveness of the model proposed in this thesis in guiding the diagnosis and practice of survival analysis.