Applications Of Change Point Selection Process Via FDR Multiple Test Method Based On The Semi-parametric Model

Change point detecting is the detection of the state change of a random process,where the change point corresponds to the moment when the probability distribution of the data stream changes,and the decision maker needs to judge whether the probability distribution of the data changes according to the sequential observation data.The advent of the era of big data has produced high dimensional data,thus how to effectively detect abnormal mutation points in the data has attracted much attention.Change point detecting technology is one of the most effective methods to solve this problem.The detection method has been widely used in many fields,such as economics,climate simulation,biomedicine,anti-terrorism security and so on,which has important research significance.This paper mainly studies the change point detecting and corresponding estimation of varying coefficient models.The change point in the data stream corresponds to the moment when the probability distribution of the data stream changes,and the decision maker needs to determine which moment the probability distribution has changed,that is,to detect the change point.At present,there are many mathematical models for change point detecting,such as mean value model,time series autoregressive model,linear regression model,etc.,corresponding to the proposed relevant change point estimation algorithm,which can effectively detect the position of strain points.But for the longitudinal data varying coefficient change point model,the general algorithm has poor identifiability.For the change point detecting problem of semi-parametric model,the traditional t-test method is set by a single p-value,which has certain limitations.Therefore,this paper constructively applies false discovery rate(FDR),an error metric in multiple hypothesis testing,to the change point detecting.Based on the relevant results of the change point detecting theory,Change point detecting method of the longitudinal data varying coefficient change point model is proposed,and its effectiveness is verified by simulation and a real data analysis.Specific contributions of this paper are summarized as follows:Firstly,this paper proposes a method for change point detecting based on FDR multiple test,and built the estimates of the parameters of longitudinal data varying coefficient change point models.Secondly,after studying the change point detecting problem of the longitudinal data varying coefficient change point model,this paper combines the found change points with the B-spline smoothing method to fit the model,and proves the large sample properties of the corresponding estimator.Thirdly,the effectiveness of the proposed method is supported by numerical simulation.Firstly,the Group Lasso method combined with non-parametric smoothing method is used to select variables,in which the regularization parameters are selected by BIC criterion.Simulation results show that the method proposed in this paper can achieve good estimation results.Fourthly,the adjusted Growth and Health Study dataset(NGHS)of the American Heart,Lung and Blood Institute was selected as the example data.Data were analyzed to determine the time-varying effects of race,height,and BMI and the newly generated covariates on the new response variables.To sum up,this paper proposes a change point detecting method based on FDR criteria for longitudinal varying coefficient change point model,so as to find the moment points of state changes in the model and take appropriate countermeasures in time,which has a wide application field.