An Empirical Analysis Of Employee Turnover Based On Data Mining

The problem of employee turnover is always the inevitable topic in enterprise management,predicting the turnover tendency of employees in advance,and taking corresponding measures,is of great significance in the management of enterprise human resources.Aiming at the problem of employee turnover,firstly this thesis uses the SMOTE method to balance the data,on the basis of which,the CART decision tree,AdaBoost and random forest model are constructed respectively,the operation of pruning and adjusting the optimal parameters is added to the model.And then the correlation between the characteristic attributes of the departing personnel and the separation is obtained.Combined with the measurement indexes of accuracy,precision,recall,F1-score and AUC value,the prediction results of three different classifiers are analyzed and compared.Finally,the prediction result is analyzed and the relevant improvement measures are put forward.The experimental results show that three kinds of classifiers have strong feasibility and reliability for employee turnover prediction,and all kinds of measurement indexes of random forest model are better than CART decision tree and AdaBoost,which is more suitable for the analysis and prediction of employee turnover problem.By sorting the importance of variable characteristics,there are 5 main factors that affect employee turnover,namely: Employee satisfaction with the company,seniority,average monthly working time,promotions in the past five years,number of projects.Combining the behavior data of employees,the random forest model is applied to the empirical evidence,which is helpful for the relevant decision makers to make reasonable decisions on employee turnover behavior.