This study investigates how machine learning can enhance traditional job evaluation frameworks used in compensation consulting. The objectives are threefold: (1) to determine how much variance in pay can be explained by common compensable factors such as education, experience, and job complexity; (2) to assess how accurately machine learning models can predict equitable pay levels compared to traditional point-factor methods; and (3) to analyze how the relative importance of these factors has shifted over the past decade. Ten years of federal workforce data (2015–2024) were integrated with O*NET occupational attributes to reconstruct a nine-factor evaluation framework. Ridge and XGBoost regression models were developed using the same compensable factors to ensure comparability. XGBoost explained 53.1% of salary variance (R2 = 0.531), outperforming the traditional framework (34.0%), and achieved 60.0% accuracy within ±20% of actual pay versus 47.4% for the traditional model. SHAP analysis revealed declining importance of education and cognitive complexity, with increasing emphasis on experience and supervisory responsibility. These findings demonstrate the potential of explainable machine learning to improve pay equity and transparency in compensation design.