Enhancing recruitment strategies with a hybrid AutoML approach to employee promotion prediction

• AutoML accelerates HR promotion prediction pipelines. • Hybrid AutoML–interpretable model achieves top performance (0.97 accuracy). • SMOTE + Tomek/ENN balance imbalanced HR datasets effectively. • Promotes fair and explainable promotion decisions in HRIS. • Merges automation efficiency with human-centered model interpretability. This research proposes a hybrid Automated Machine Learning (AutoML) architecture for predicting employee promotions within Human Resource (HR) analytics to address the ongoing issue of class imbalance. A series of traditional machine-learning models and AutoML pipelines were evaluated comparatively alongside multiple resampling methods including SMOTE, SMOTE + Tomek Links, and SMOTE + ENN. In the proposed framework, the TPOT AutoML optimizer was integrated with interpretable learners, K-Nearest Neighbors and Extra Trees Classifier, through soft voting, which balances automation, accuracy, and interpretation. Results demonstrate that strong preprocessing methods and hybrid optimization methods improve predictive performance, achieving accuracy of up to 0.97 and F1 scores of 0.94 under SMOTE + ENN. Although ensemble models, such as Random Forest and LightGBM, performed equally well, the hybrid AutoML approach yielded better minority-class recall across the dataset. In addition to aggregate metrics, the study also illustrates potential HRIS-oriented deployment scenarios for explainable predictive analytics under controlled experimental settings. Despite the unavailability of actual HR data based on confidentiality, validated open data and cross-validation supported reproducibility and ethical considerations. This study presents an empirical evaluation of hybrid AutoML pipelines for promotion prediction under class imbalance, emphasizing integration rather than algorithmic novelty. Results are reported on a publicly available HR dataset and should be interpreted as illustrative of pipeline behavior under controlled conditions. The contribution lies in demonstrating how AutoML and interpretable ensemble models can be combined to improve robustness and minority-class recall, providing a reproducible baseline for future HR analytics research.