PERFORMANCE ANALYSIS OF MACHINE LEARNING BASED CLASSIFIERS FOR PREDICTING EMPLOYEE ATTRITION IN RESOURCE-CONSTRAINED HR SYSTEMS

Abstract

The problem of employee turnover is getting worse for many tech companies throughout the world. Staff turnover, sometimes known as attrition, is on the rise and particularly high in the information technology sector for reasons that are difficult to predict or predictably address. Any industry, no matter how big or little, can benefit from this. Keeping an existing employee costs much less than recruiting a new one. Acquiring a new resource necessitates a period of adjustment to a new team's or company's culture as well as training and the transfer of existing knowledge. Using the IBM HR dataset obtained from Kaggle, this project endeavors to forecast employee turnover in HR systems with limited resources. Data cleaning, encoding, normalization, feature extraction, and class balancing using SMOTE were all steps in the extensive pretreatment pipeline that was applied to the dataset, which contains 1,470 employee records with 35 features. Evaluate the predictive power of various ML models, including LR, RF, SVM, and a proposed DNN. The DNN model achieved the best results of all the models when it came to capturing complex patterns in the data. It achieved an F1-score of 94.52% and had very high accuracy and precision. Using deep learning approaches to enhance HR decision-making in environments with limited resources is essential for employee retention.