Outlier Detection Schemes For Profiles With Binary Data

Statistical Process Control(SPC)has received sufficient attention since the concept of Control Chart was first proposed by Dr.Shewhart in the 1920 s.With the progress of science and technology,large quantities of problems of product quality control emerge in the process of flourishing development of manufacturing industry.The appearance of these problems also promotes the perfection of academic research in the field of SPC.In many real-world applications,quality attributes are no longer simple,unary or multivariate variables,but rather are characterized by the relationship between a response variable and one or more explanatory variables,which is tipically referred to as a profile.Recently,profile monitoring,as an important research topic in SPC,has been included in the scope of their research by more and more experts and scholars.Most of the previous researches on profile control are based on the basic assumption that the response variable is continuous and subject to normal distribution.However,the profile whose response variable is categorical data is ubiquitous but not discussed enough in the relevant literature.In this thesis,we focus on the profile process with binary response data,using the variable selection method based on the penalized likelihood,so as to propose systematic outlier detection schemes for the Phase-I control.To make the research clearer,a combination of the simulation and the application of a real example is utilized.An important issue to be aware of when detecting outliers is the masking effect.The masking effect caused by the failure to fully detect outliers can seriously distort the estimation of the profile model.The simulation results show that compared with the existing control chart,the recommended scheme in this thesis has higher detection efficiency in practice and can detect all the outliers in the dataset with higher probabilities.Moreover,the masking effect is also smaller.Finally,this thesis describes the operation procedure of the proposed scheme through a real example application of the automobile warranty claim data.