<![CDATA[This study aims to understand and predict the unemployment rate patterns based on educational background in Indonesia between 1986 and 2024, with a focus on university graduates. The data, which was initially complex, was successfully processed into a format ready for time series analysis, and the GARCH (Generalized Autoregressive Conditional Heteroskedasticity) model was applied to measure the volatility of unemployment. The evaluation results show that the GARCH model's assumption regarding the stability of the average unemployment rate is inaccurate, as evidenced by the large error values (RMSE 283209.26 and MAE 246252.37), indicating that this model does not fully capture the fluctuations in unemployment. The average coefficient (mu) is 436.50, and the log-likelihood is -284.05, with conditional volatility forecast values ranging from approximately 1.91e+11 to 2.79e+11. The Bayesian Structural Time Series (BSTS) model was also applied to decompose the data into long-term trend components and seasonal patterns, providing a clearer picture of unemployment movement. However, technical constraints in the implementation of BSTS using TensorFlow Probability resulted in predictions not being completed. Nevertheless, this analysis shows that the unemployment rate of university graduates is highly volatile, and improvements in the GARCH model, as well as resolution of the technical constraints in the BSTS model, are crucial for generating more accurate and reliable predictions.]]>
Copyrights © 2025
