AN EFFICIENT SUPERVISED MACHINE LEARNING MODEL USED FOR CLASSIFICATION AND PREDICTION OF EMPLOYEE ATTRITION

Abstract

Abstract—A company's employees are its most important asset. It is crucial to know who could quit the company due to the high expense of professional training, the loyalty that has grown over the years, and the delicate nature of some organizational roles. Employee attrition (EA) can occur for several causes. Employees leave a company either voluntarily or involuntarily; this phenomenon is known as attrition. Researchers must investigate the use of machine learning (ML) in corporate organizations due to the increasing fascination with the topic among corporate decision-makers and executives. The use of ML models to investigate employee turnover (ET) is the focus of this study. Using data analysis and ML approaches, this research primarily aims to explain a methodical flow for forecasting attrition. Acquiring data, preprocessing it, visualizing it, engineering features, balancing it, and finally classifying it using a ML approach are the stages involved. Automated and precise prediction of EA is achieved in this work through the use of supervised ML models. The article details the use of k-best for feature engineering and ML classification techniques such as Random Forests (RFs) and decision trees (DTs) to forecast the likelihood of ET for each given new hire. Using the Kaggle dataset titled "IBM HR analytics employee Attrition Performance," this article trains and evaluates a RF model with cross-validation in a Python environment.  The ultimate goal of effectively identifying attrition is to help any company enhancevarious retention tactics on important workers and raise the pleasure of those employees. Classification predictions were evaluated using the following 5 efficiency indicators: f1-score, Accuracy, Confusion Matrix, Precision, Recall,  & the ROC Curve. The findings showed that the RF classifier had the best performance, with an accuracy of 96%.  Businesses' capacity to avoid ET might be significantly affected by the results of this study.