Statistical Analysis And Prediction Of Employee Turnover Intention

In recent years,with the continuous improvement of China's comprehensive national strength,continuous progress in science and technology,and steady progress in opening to the outside world,Chinese enterprises have become stronger and stronger.Currently,among the fortune 500 companies,there are 129 from China.Talent is the core component of an enterprise's competitiveness,and the cultivation and introduction of talents is the top priority of every enterprise.The competition for talents promotes the increase of the mobility of employees,which makes it a common phenomenon for employees to leave their jobs.The increase of employee turnover rate in an enterprise will not only increase the cost of human resources,but also have a negative impact on the enterprise's reputation,thus increasing the difficulty of talent introduction.Therefore,to stabilize and attract talents,to analyze the factors leading to the dimission of employees,and to predict and control the dimission tendency of employees are important issues concerned by all enterprises.Based on the real data released by Kaggle platform in 2017 in the employee turnover prediction training competition,this paper USES R and SPSS software for statistical analysis.In the study of the influencing factors of employee didication,the dependent variables involved are often presented as dichotomous variables,so it is more appropriate to use generalized linear model for analysis.In the classification and prediction of variables,the use of decision tree for analysis has its unique features,which is convenient to realize and understand.Compared with decision trees,random forest is more advantageous in processing large amounts of data.However,these two algorithms sometimes have limitations,which make the analysis results relatively simple and prone to overfitting.For this problem,the XGBoost algorithm adopted in this paper has more advantages.At the same time,the prediction results of CART regression tree and random forest in decision tree are given to make model comparison.Using the above human resources data,this paper focuses on the study of whether the enterprise employees dimission,the probability of dimission and the main factors leading to dimission.Before the discussion,firstly,the research background and purpose of the problem were briefly described.Secondly,a lot of energy was spent on data preprocessing,quantification and screening to make it meet therequirements of modeling.Then the relevant knowledge and statistical analysis tools are introduced,such as the basic principles of various algorithms and the analysis process.Next,Logistic regression,Probit model,multinomial distribution log-linear model,CART regression tree,random forest and XGBoost algorithm were used to conduct empirical analysis on the data,and the six analysis results were compared and predicted by modeling,testing and optimization.The empirical analysis shows that,compared with Logistic model and Probit model,the fitting degree of multinomial distribution log-linear model is higher.And the influence factors of employee turnover in general are not affected by a single variable,even if the dependent variable and the single variable present positive correlation,but in the case of add more influencing factors,the result is often change,may produce and single variable case different results,such as irrelevant,or negative correlation.By comparing the prediction accuracy of the six algorithm models,the XGBoost model performed best,followed by random forest and decision tree,because the regularization method was used to prevent overfitting and the second order Taylor expansion was used to make loss control more accurate.A new conclusion of this study is that employee turnover is not determined by a single variable,but by multiple factors,which is an important supplement to previous research results.In addition,this paper USES six models for prediction comparison,screening out the relatively better model.In a word,the research work of this paper is helpful for the further discussion in the future...