PREDIKSI VOLUME IMPOR KEDELAI INDONESIA MENGGUNAKAN EXTREME LEARNING MACHINE (ELM): PENDEKATAN MACHINE LEARNING DALAM ANALISIS PERDAGANGAN INTERNASIONAL

Abstract

Indonesia faces structural vulnerability in food security, driven by a persistent supply-demand gap in strategic agricultural commodities, particularly soybeans. Despite extensive trade policy research, existing literature predominantly relies on linear econometric models that fail to capture non-linear dynamics and non-stationary pattern shifts in macroeconomic time-series data. To address this methodological research gap, this study applies the Extreme Learning Machine (ELM)—a single-hidden-layer feedforward neural network selected for its universal approximation capability, analytical parameter solving without backpropagation iterations, and avoidance of local minima traps—to predict Indonesia’s soybean import volume for 2026–2035 and determine key macroeconomic drivers. Using secondary annual time-series data (2010–2025) from BPS, Bank Indonesia, FRED/IMF, and the Ministry of Agriculture, four independent variables were evaluated: USD/IDR exchange rate, world soybean price, domestic production, and national consumption. Feature importance was assessed via Pearson correlation, Random Forest, and Ridge Regression. The optimized ELM model (10 hidden neurons, sigmoid activation) achieved exceptional predictive accuracy with R² = 0.9519 and MAPE = 2.75% on training data, and a robust Leave-One-Out Cross-Validation (LOOCV) MAPE of 8.70%. Baseline comparisons confirmed ELM outperforming traditional ARIMA (MAPE = 14.32%) and Multiple Linear Regression (MAPE = 12.18%). Exchange rate and national consumption emerged as dominant predictors. Projections indicate a declining import trend from 2.34 million tons (2026) to 0.88 million tons (2035), representing an import squeeze effect under exchange rate pressure rather than domestic self-sufficiency. The primary scientific novelty lies in establishing a non-linear machine learning framework for food import forecasting in emerging market economies, providing actionable quantitative evidence for exchange rate hedging and agricultural trade policy.