Optimizing Random Forest Networks for Human Resource Skill Gap Analysis Based on Future Workforce Demands

Abstract

The accelerating impact of artificial intelligence, automation, and digitalization has disrupted traditional labor markets and created urgent challenges in aligning workforce skills with future demands. This problem arises because existing forecasting methods, such as ARIMA or conventional machine learning approaches, often fail to capture the nonlinear and temporal complexity of skill dynamics, leading to inaccurate predictions of future shortages and surpluses. This study aims to develop a predictive framework that improves accuracy in skill gap analysis and provides actionable insights for workforce planning. The proposed model, named Optimized Random Forest Network (RFN), integrates heterogeneous data sources including online job postings, occupational taxonomies (O*NET), and macroeconomic indicators. The model incorporates temporal feature extraction, text embedding of job descriptions, and exogenous signal integration, combined with hyperparameter optimization and ensemble refinement to strengthen robustness. The results demonstrate that the optimized RFN outperforms baseline models such as standard Random Forest, Gradient Boosting, and ARIMA achieving superior performance in regression (sMAPE = 9.1%) and classification tasks (Macro-F1 = 0.82). Furthermore, the analysis highlights increasing demand for skills in data analytics, artificial intelligence, and green technologies, while routine-based roles show declining relevance. These findings offer valuable contributions for policymakers, industries, and educational institutions in designing adaptive strategies to bridge skill gaps and align human resources with future workforce demands.