Predicting Indonesia's Unemployment Rate Using Machine Learning and Macroeconomic Indicators: A Time Series Approach

Unemployment is a significant global challenge with profound socio-economic impacts that affect individual well-being, increase inequality, and hamper national economic stability. For Indonesia, as the largest economy in Southeast Asia, understanding the complex dynamics affecting unemployment and developing robust forecasting methods are crucial for effective policy formulation. This study aims to use machine learning (ML) techniques to predict the unemployment rate in Indonesia by analysing annual data from 1970 to 2023, including important factors like Gross Domestic Product (GDP) growth, higher education levels, inflation, and foreign direct investment (FDI). Eight ML regression models were evaluated using a time-based holdout approach (1970–2022 for training, 2023 for validation). Model performance was assessed using Mean Absolute Error (MAE), Mean Squared Error (MSE), R², and Mean Absolute Percentage Error (MAPE) on validation data. The ensemble model, particularly the Gradient Boosting Regressor, performed the best in recognising past trends (R² 99% on training data) and had the highest accuracy for the 2023 data (predicted 3.79% compared to actual 3.308%, MAPE 14.7%). Feature importance analysis of the best model identified higher education, GDP growth, and FDI as the most influential drivers of the unemployment rate. Using the best gradient boosting model and independent variable forecasts (via univariate ARIMA model), Indonesia's unemployment rate is predicted to experience moderate fluctuations from 2024 to 2028. This study shows that using advanced machine learning techniques can accurately predict unemployment in Indonesia, highlights the main factors affecting unemployment, and provides a strong model that can help create policies to stabilise the job market and lower unemployment.