Forecasting the unemployment rate in West Java Province using VARX and SVR methods

Abstract

This study discusses the forecasting of the Open Unemployment Rate (OUR) in West Java Province using two time series approaches: Vector Autoregressive with Exogenous variables (VARX) and Support Vector Regression (SVR). The dataset consists of monthly observations from 2018 to 2023, including variables such as OUR, the Labor Force Participation Rate (LFPR), Gross Regional Domestic Product (GRDP), and the Human Development Index (HDI). Based on the optimal lag selection using the AIC, the VARX model produced the best lag configuration of (5,2), consisting of five lags for endogenous variables and two for exogenous variables. Meanwhile, the SVR model was developed through Grid Search to find the best parameter combination, resulting in a linear kernel with  and . The evaluation results showed that the SVR model performed better than VARX, with MSE, RMSE, and MAPE values of 0.24, 0.49, and 6%, respectively, lower than those of the VARX model, which reached 0.68, 0.82, and 8.4%. SVR was selected as the best model and used to forecast the OUR until the end of 2025. The forecast results indicated a spike in OUR at the beginning of 2024 at 8.52%, followed by a declining trend that continues and stabilizes in the range of 7.96%-8.12% by the end of 2025. In conclusion, SVR outperforms VARX in predictive accuracy, while VARX remains useful for analyzing inter-variable relationships.