Group Detection Based On Variable Selection

As a means of medical testing,group testing saves a lot of money and time compared to individual testing.”Whether an individual is sick or not is related to many factors.Establishing a model based on related factors to speculate whether an individual is sick or not is our main issue in this article.".With the gradual increase of related factors,some unnecessary covariate information affects the fitting effect of the model,and may not contribute to the results of the model.Therefore,we need to select the covariate information related to the dependent variable through known data information.Grouping testing involves collecting individual specimens,such as blood,urine,swabs,etc.,and detecting the presence of diseases in these samples.When individual covariate information is available,such as age,gender,number of sexual partners,etc.,a common goal is to associate an individual’s true disease status with the covariates in the regression model.How to establish the relationship between these two through a model is the most fundamental problem in group detection methods,as no real individual state is observed,and all test values are prone to misclassification due to measurement errors.The regression methods previously used for group testing data may have been inefficient because they were limited to using only the initial group state or they made unrealistic assumptions about the probability of measurement accuracy.To overcome these limitations,this article proposes a new method for solving variable selection in high-dimensional models in group detection.The method consists of two parts.Firstly,the Bayesian method is used to determine the probability of each covariate being included in the model,and then the maximum a posteriori estimation algorithm is used,combined with the expectation maximization algorithm to calculate the model parameters.The article proves the reliability of our proposed method from both theoretical and experimental simulations.Compared to existing Bayesian methods that only extend to linear models and generalized linear models,our method is novel in extending edge inclusion probability to group detection.