Application Of Classifier In Employee Turnover Prediction

Based on the decision tree classification technology in data mining,this paper establishes a combined classifier model to classify the employee turnover data set and predict the employee turnover.The main contents of this paper include data preprocessing,single classifier model establishment and model optimization through pruning,combination and other means to improve the accuracy of prediction.In this paper,data preprocessing,model building and other data mining methods are used for employee turnover data set.The method of data discretization and redundant attribute deletion is used to preprocess the data and optimize the data set.After that,the problem of unbalanced data set is overcome by using the method of put-back sampling.Compared with ID3 algorithm and C4.5 algorithm,a single decision tree classifier is built on this data set,and then a combined classifier is built on the basis of a single decision tree by using the combined classifier bagging algorithm to improve the prediction efficiency,predict employee turnover and evaluate the enterprise employment risk.Specifically,this paper research content mainly has the following several aspects:(1)Preprocess employee turnover data.Discretization of continuous data for model processing,filling or deleting samples with incomplete attributes.Then 3 methods are used to remove redundant attributes to simplify the data set.(2)Deal with data imbalance.The data set used in this paper has a positive and negative sample ratio of 1:10,which is a typical unbalanced data set.The data set is processed into a balanced data set using a sampling method so that it can be used by the classical model.(3)Establish a single decision tree model.Comparing two classical decision tree classification algorithms in data mining,observing the efficiency and accuracy.Then the decision tree model is improved,pruned and visualized.(4)Establish the combinatorial classifier.The combination classifier constructed by bagging algorithm was used to observe the accuracy,and the weighted voting method was used to obtain the prediction results.By comparing the experimental results,the following conclusions can be obtained:During the data preprocessing stage,removing redundant attributes can reduce the model construction time.But the accuracy doesn't change much.This shows that the above method can effectively reduce the dimension of the data set.Finally,in the stage of model construction and classification prediction,the decision tree model can be effectively simplified by pruning operation.And improvement and combination can improve the accuracy of prediction.