| With the deepening development of the market economy,the flow of employees among enterprises has become normal,and the phenomenon of employee loss is inevitable.If the proportion of employee turnover is within a reasonable range,it will not affect the normal operation of the enterprise,but if the proportion of employee turnover is beyond the range that the enterprise can bear,it will directly affect the normal operation of the enterprise.Only when the staff team is stable can an enterprise be stable and develop towards the established goals.The purpose of studying employee turnover is to make enterprises do a good job of early warning,prepare solutions for employee turnover,improve the human resource management system,develop human resources at a lower cost and increase enterprise output.Therefore,it is a long-term and very important work to study the employee turnover in enterprises.In the digital and under the background of big data technology mature gradually,from the perspective of human resource management,can collect data related to employees work,combined with relevant theory of human resource management potential information mining data,build reasonable risk model,make enterprise can better understand the needs of the employees,to evade risks related by the departure of employees,to better manage the staff turnover.In this paper,the survival analysis model is introduced.Firstly,the life table analysis and Kaplan-Meier estimation are used to make a descriptive analysis.The purpose is to comprehensively understand the structure of sample data and the time distribution of employee dimission.Then,Cox model is constructed,stepwise regression method and Lasso method are used to select the variables that have significant influence on employee dimission,and the relevant quantitative analysis is made.Finally,an accelerated failure time model was built.Based on the Cox model,15 important covariables influencing employee diission were selected and four accelerated failure time models,namely,Weibull distribution,exponential distribution,lognormal distribution and loglogistic distribution,were fitted.And by AIC criterion,the accelerated failure time model with T following Weibull distribution fitting has the best effect.With the popularity of machine learning,many machine learning models such as RF,SVM and KNN have been applied to the prediction of employee turnover.But most forecasts are based on categorization--determining whether an employee will leave or not--rather than analysis of the risk factors that influence employee turnover and estimates of how long the employee will remain with the company.This paper studies the risk of employee turnover,which can solve two problems.First,the construction of Cox model can make the risk factors have a good explanation;Second,an accelerated time-of-failure model can be built to estimate how long employees will remain with the company. |
