Forecasting Employee Attrition in HRM Based on Advanced Supervised Machine Learning Models

Abstract

Human resources are a company's most valuable asset, so losing employees is a big deal for the business. Human resources managers and department heads have a number of challenges when trying to predict when employees may start to leave their positions. Consequently, this research provides a machine learning-based methodology for HR planning-related employee turnover prediction utilizing the IBM HR dataset. Data cleaning, label encoding, min-max scaling, and class balancing with SMOTE were all part of the preprocessing stages to ensure reliability and reduce bias. Including 35 demographic, job-related, and satisfaction-related factors, the dataset comprised 1,470 employee records. Decision Tree, AdaBoost, and Support Vector Machine were some of the more conventional models tested with two more modern ensemble models, XGBoost and Random Forest (RF). As shown in the testing findings, XGBoost achieved a 96.95% accuracy rate and RF a 93.90% accuracy rate, both of which were better than the benchmark models. With its outstanding generalizability and predictive capacity, XGBoost stood out as the most effective of the bunch. The results show that the suggested models can handle complicated, high-dimensional HR data and provide dependable predictions of employee turnover, demonstrating their resilience and scalability. By allowing firms to detect people at risk early and put effective retention initiatives in place, this adds to proactive HR decision-making.