Particle swarm optimization-based support vector regression for unemployment rate prediction using panel data

Abstract

Predicting the Open Unemployment Rate (OUR) is important for supporting data-driven employment policies, particularly in regions with complex economic and social conditions. This study aims to develop a predictive model for OUR using panel data from districts/cities in West Java Province during the 2018–2025 period by applying Support Vector Regression (SVR) optimized with Particle Swarm Optimization (PSO). The dataset includes economic, social, and demographic variables, namely labor force participation rate, average years of schooling, population, minimum wage, Human Development Index, GRDP per capita, poverty rate, and population density. The proposed approach combines SVR as a nonlinear regression technique with PSO for hyperparameter optimization to improve prediction accuracy. The experimental results show that the model achieved a Mean Squared Error (MSE) of 1.4888 and a coefficient of determination (R²) of 0.4343, indicating moderate predictive performance. In addition, the optimization process demonstrated a stable reduction in RMSE values during iterations, confirming the effectiveness of PSO in enhancing the SVR model. The findings suggest that the SVR–PSO model is capable of capturing general unemployment patterns in panel data and can support adaptive, data-driven employment policy analysis.