Research And Implementation Of Lung Prediction Model Based On Particle Filter For Parameter Estimation

The incidence rate and death rate of carcinoma of the lungs patients are gradual upward trend,carcinoma of the lungs has become a malignant tumor sickness menaceing human vita and health.In the study of survival prediction model for lung cancer patients,when the model reaches a certain sample size,the utilization rate of characteristic data of follow-up patients is reduced,and the accurate prediction is difficult to improve.The research goal of this thesis is to establish a prediction model,so that the model can update the model parameters according to the new patient data and improve the accuracy of the model.The design of the prediction model can provide some reference and basis for differential treatment of lung cancer patients after surgery,improve the intervention effect of treatment and further improve patient survival.This thesis mainly focuses on: 1)the key characteristic variables are determined.According to the principle of explaining the model as highly as possible with as few variables as possible.The main component analysis and combination of LASSO regression and random forest are compared.Finally,the method of combination of LASSO regression and random forest is selected for dimension reduction,The 70 feature variables in the original data were reduced to 5characteristic variables which were highly correlated with the survival of patients,which solved the issue of direct modeling model over fitting;2)The predictive survival model of carcinoma of the lungs patients is builded,and the risk characteristics obtained by feature reduction were changed into dumb variables,and then the logistic regression prediction model was established.The regression coefficient was tested significantly.The results showed that the prediction accuracy was 90%,but the recall rate of the dead was 60%,The accuracy is difficult to improve with the increase of data;3)This paper realizes the combination of parameter estimation and data assimilation.It takes the coefficient and the intercept in regression as parameter The idea of particle filter is used to make the model parameters complete the prediction and update operation with the continuous addition of new feature data.The overall accuracy,recall rate and AUC value of the prediction model are improved on the same test set,and the overall accuracy is improved to92%,the recall rate of dead patients increased to 71%.The innovations of this thesis are as follows: 1)From the model point of view,the combination of particle filter and parameter estimation is realized.In particular,the coefficients and interceptions in the logistical regression are taken as the parameters to be estimated and the parameter process as the only objective state of the filter issue,and then the measurement equation in the filtering process is constructed.Through the continuous introduction of data,the prediction and update process in the particle filter algorithm process can be completed,and the parameter estimation value can be obtained,which further improves the accuracy and recall rate of the prediction model.2)From the apply of angle,on the one hand,it can make better use of the characteristic data of follow-up patients,on the other hand,it can make the relevance of different patients stronger,so as to achieve precision medicine.