Choosing machine learning model for predicting employee churn in the telecommunication industry

One of the biggest problems of businesses today is losing their employees. Employee churn has many costs to companies. For this reason, it is very important to predict the loss of employees and take precautions. The accuracy of the estimates made on such an important issue is also very important to ensure that the actions to be taken are not erroneous and to reduce the churn of employees. There are many estimation methods, but in this study, an employee data set belonging to the telecommunications sector was analyzed by using the classification method, which is one of the machine learning methods. The aim of the study is to analyze the data set with eight classification models and to propose the most suitable classification model for this problem. These models are coded with Python language. 70% of the dataset was used in training and validation the model and 30% in testing the model. The applied models were evaluated according to accuracy, cross validation score, precision, sensitivity, 𝑓𝑓1 score and Area Under the Curve (AUC) metrics. Among the models used, the best classification model was found to be the random forest model with an accuracy of 92.2%. The second best model was found to be the gradient increasing machines model with an accuracy value of 91.4%. k nearest neighbor is the worst classifying model among the applied models with an accuracy rate of 89.1%. When the models applied in this study are evaluated for the classification studies to be carried out in the future, the random forest model, which has the best metric values, that is, which classifies the best, is recommended.