Mixture Models For Clustering With Censored Data And Its Decision Support In Logistics

Censored data widely exists in engineering practice and management,and it presents new challenges to data-driven logistics decision-making methods and theories.Censored data is difficult to deal with traditional decision-making methods and models due to the particularity and complexity of its censoring mechanism.Therefore,based on the censorship mode and data distribution of censored data,this paper uses the theory of expectation maximization algorithm,zero-inflated Poisson mixed model,likelihood function adjustment and model verification criteria to study the research on hybrid clustering model and its application in logistics decision support.The main work and innovations are as follows:(1)Comprehensively reviewed related research on censored data,and defined censored data and its likelihood function.From the censorship mode and dimension,the adjustment of the likelihood function of the censored data is analyzed.(2)The expectation maximization algorithm of censored data following Gaussian mixture distribution is studied.First,a Gaussian mixture clustering model based on the standard expectation maximization algorithm based on random missing data is given,and the model parameter estimates and likelihood functions are derived.Then,a Gaussian mixture clustering model based on the censored data's expectation maximization algorithm is given.After analyzing the censored data,the adjusted likelihood function and the full data score vector are used to infer the parameters of the censored data.estimated value.The clustering effect and verification criterion prove that the model of censored data proposed in this paper is better.(3)The expectation maximization algorithm of censored data obeying Poisson mixed distribution is studied.First,the relationship between Poisson distribution and Gaussian distribution is discussed,which leads to the problem of Poisson censorship under the queuing theory model.Then based on the idea of a Zero-Inflated Poisson mixed model,a Scale Inflated Poisson distribution model is proposed.The model also uses the expectation-maximization algorithm to calculate the model parameters.The difference is that the model can estimate the censored data distribution based on the parameter estimates.This model breaks through the data limit of the Zero-Inflated Poisson model and is suitable for more complex censored count data.(4)Data-driven logistics decisions require high-quality data.Whether it is new retail,smart logistics or storage center capacity planning and other data-driven logistics projects,cannot do without data support,the importance of data construction can be seen.Simulation experiments and application cases have proved that the method proposed in this paper can improve data quality,correct parameter estimation and data distribution,and provide new ideas and theoretical support for logistics decision-making from the perspective of data quality.