Stacking-Based Transfer Learning for Optimized Employee Management and Retention in HR Systems

Background: For human resources departments trying to ensure organisational stability and lower turnover costs, employee retention is a major challenge. In terms of employee management, traditional approaches often lack the capacity to correctly mix insights from several sources, so limiting the predictive power of these approaches. Methodology: This work proposes a Stacking-Based Transfer Learning (TL) model to support bettering of employee retention policies. This model aggregates domain-specific data and pre-trained models to offer the most exact projections. The approach extracts and combines textual and numerical insights from HR data by means of a hierarchical stacking ensemble of deep learning models, which consists of BERT and ResNet. Tests carried out on a dataset of employee retention easily accessible to the general public reveal the efficiency of the model. Findings and Result: With 92.5%, the suggested method's accuracy exceeded those of other now in use methods including Random Forest (88.3%) and Support Vector Machine (85.7%). Moreover, by 17% over baseline models, the Stacking-based TL model lowers the prediction error linked with employee turnover rates. These findings indicate that HR analytics has great potential to include innovative learning strategies into order to increase employee satisfaction and retention rates respectively.