Prediction of Employee Turnover with Imbalance Dataset Using Machine Learning Methods

Abstract

Employee turnover can have a significant impact on an organisation's productivity, culture and profitability. Accurately predicting employee turnover can help organisations proactively identify and address issues before they become major problems. In this paper, various analyses were performed with the help of traditional machine learning methods using employee turnover and attrition dataset. As a result of the analyses, unbalanced data distribution was detected in the dataset. In order to solve this problem, methods for balancing up and down data sets were used. After data balancing, the k-fold method, one of the cross-validation methods, was applied to avoid overlearning. The Random Forest Classification method was selected and used together with the ROS method, which shows higher performance. GridSearchCV, a hyper-parameterisation technique, was applied to the selected model to select the best parameters. At the same time, both data pre-processing and postprocessing activities were performed. As a result of the experiments conducted in the study, it was found that the data set balanced using the proposed method increased the performance values in the classification result and improved the classification performance compared to the raw data set and other sampling methods.