SENIORITY, TENURE, AND WAGES IN THE IT INDUSTRY: A 2024 ANALYSIS USING MACHINE LEARNING APPROACHES

Abstract

This study models the determinants of IT professionals’ wages in 2024 and translates the findings into actionable guidance for managers. A standardized data pipeline (deduplication, outlier screening, consistent handling of missing values, principled encoding of categorical variables, and normalization of continuous predictors) preceded estimation of a LightGBM model with stratified cross-validation. The model achieved R² = 0.65, indicating an adequate balance between explanatory power and practical utility for compensation design. The resulting influence map shows seniority/grade and years of experience as the dominant contributors to explained variation. Secondary but meaningful effects arise from job role, English proficiency, primary programming language, and company type; region and company size play smaller roles, while formal higher education contributes minimally once professional attributes are controlled. Stability checks-subgroup analyses, bootstrap, and permutation tests-confirm the robustness of factor rankings and error metrics. Managerial significance: The results provide a defensible basis to calibrate transparent pay bands by role and grade, link promotions to verifiable expansions of responsibility, prioritize targeted investments (e.g., functional English training), and align remote/hybrid policies with role-specific coordination needs-enabling consistent, fair, and budget-aligned compensation decisions. Organizations can operationalize the findings via a quarterly compensation governance cadence that audits for title inflation and pay compression against model-aligned targets. Local model explanations can be embedded into 1:1 development plans to set promotion readiness criteria and to communicate clear, evidence-based pathways for progression. Limitations include the cross-sectional, self-reported nature of the data and potential unobserved heterogeneity; future research should incorporate panel designs and quasi-experimental validation where feasible.