What question did this study set out to answer?

The central aim is to enhance prediction accuracy for employee attrition using a hybrid machine learning framework.

April 4, 2026Open Access

Mitigating Class Imbalance in Human Resources Attrition Prediction using Hybrid Generative and Cluster-Driven Ensemble Modeling

Key Points

The central aim is to enhance prediction accuracy for employee attrition using a hybrid machine learning framework.
Utilized K-Prototypes clustering for employee segmentation based on both numerical and categorical data.
Applied Conditional Tabular GAN to generate synthetic samples to address class imbalance.
Developed a stacking ensemble model incorporating base learners like ANN, XGB, and CatBoost for better predictions.
The hybrid framework achieved higher accuracy compared to standard models.
The recall rate and AUC-ROC were significantly improved over traditional oversampling methods.
Clustering, GAN-driven augmentation, and ensemble learning provide reliable attrition predictions.

Abstract

Modern organizations continue to have a major problem with employee attrition, which has an impact on long-term strategic planning, operational effectiveness, and workforce stability. For data-driven human resource (HR) management, it is important to be able to accurately estimate attrition. This research presents a comprehensive machine learning framework that incorporates K-Prototypes clustering for employee segmentation, Conditional Tabular GAN (CTGAN) to address class imbalance, and a stacking ensemble model for the reliable prediction of employee attrition utilizing the IBM HR Analytics dataset. K-Prototypes clustering was used to find hidden groupings of employees by modeling both numerical and categorical features together. This let us learn more about subpopulations that are likely to leave the company. CTGAN was used to make high-quality synthetic minority samples to fix the dataset’s very unbalanced character. This made the training distribution more balanced and useful. We created a stacking ensemble model with base learners like ANN, XGB, and CatBoost to find non-linear correlations and make predictions better. The experimental results show that the proposed hybrid framework is more accurate, has a higher recall rate, and has a higher AUC-ROC than standard models and traditional oversampling methods. The results show that using clustering-based segmentation, GAN-driven data augmentation, and ensemble learning together is a good way to make attrition prediction systems that are both reliable and easy to understand.

Ask AI

Helpful

Bookmark

View Full Paper