Improving Palm Oil Price Forecasting Using a Gaussian-Enhanced Generative Adversarial Imputation Network

Accurate forecasting of palm oil prices plays a vital role in guiding trade policies and maintaining market stability, especially in the growing economic and climate-related uncertainties. However, missing values in palm oil datasets present a major challenge to build reliable forecasting models. This study proposes an improvement to the Generative Adversarial Imputation Network (GAIN) by replacing its original uniform random noise generator with a Gaussian-based generator using the Box-Muller transform to improve the quality and stability of imputing missing values in multivariate datasets. The enhanced GAIN shows better imputation performance, as indicated by lower Root Mean Square Error (RMSE) and Euclidean Distance (ED) compared to the original version. This improvement also translates into better forecasting results when used with Long Short-Term Memory (LSTM), Random Forest (RF), and Linear Regression (LR) models, which achieved RMSEs of 818.45, 932.12, and 858.18, respectively. Further analysis using Pearson's correlation shows that accurately preserving key statistical properties - such as the mean and standard deviation - plays an important role in improving predictive accuracy. Overall, the proposed method offers an effective solution for handling missing data in complex, real-world forecasting tasks. Future work may explore extending this approach to capture cross-sectional patterns in panel data, enabling more context-aware and structure-sensitive imputation.