Construction and Application of Machine Learning Models for Salary Evaluation

Abstract

In the contemporary knowledge economy, data-driven salary evaluation is a strategic imperative for talent acquisition, retention, and equitable compensation. This paper presents a systematic review of the construction and application of machine learning models for salary prediction, charting the methodological evolution from the restrictive assumptions of traditional econometric paradigms to the superior predictive power of modern algorithms. To provide a holistic analysis, this study introduces a novel five-dimensional framework—”DataFeature-Model-Explanation-Governance”—that moves beyond model accuracy to encompass the entire system lifecycle. The paper reviews state-of-the-art supervised learning models, including tree-based ensembles (e.g., XGBoost) and deep learning architectures (e.g., TabNet), and examines their diverse applications in empowering individuals, optimizing corporate HR strategies, and informing macroeconomic policy. A core finding is that the central challenge in the field has shifted from the pursuit of predictive accuracy to the imperative of building trustworthy AI systems. Consequently, the paper critically analyzes the indispensable roles of data governance, model interpretability (e.g., SHAP), and algorithmic fairness in mitigating bias and ensuring responsible deployment. It concludes by synthesizing these insights into a practical roadmap for researchers and practitioners, aiming to foster the development of transparent, equitable, and intelligent compensation systems.