Explainable AI and robust forecasting of global salary trends: Addressing data drift and unseen categories with tree-based models

This article studies salary prediction under distributional drift using explainable boosting models and hybrid forecasting.  We integrate unseen-aware feature engineering, robust objectives, SHAP-based interpretability, drift detection, and time-series forecasting (Prophet/SARIMAX) on multi-year data (2020–2024), and report a comprehensive evaluation aligned with typical MMC guidelines.  Modern salary data are heterogeneous, heavy-tailed, and non-stationary.  Therefore we combine robust tree-based learners with drift monitoring and explainable forecasting to prioritize stable absolute error, transparency, and maintainability over raw variance capture.  Our best integrated pipeline reaches $R^2=0.31$ on a 2024 hold-out while keeping MAE/RMSE stable across folds, and uncovers year-to-year drift that necessitates periodic retraining monthly and quarterly forecasts indicate a sustained upward trend with seasonality, where SARIMAX captures short-term fluctuations and Prophet yields interpretable trend decompositions.