Employing Machine Learning Techniques to Forecast Food Inflation in Indonesia: The Role of Food Price Indices

Abstract

This study investigates the use of machine learning techniques, specifically XGBoost and LSTM, to forecast food inflation in Indonesia, leveraging daily food price data from the Pusat Informasi Harga Pangan Strategis (PIHPS). The research reveals that the XGBoost model significantly outperforms traditional models like SARIMA, especially when food prices and indices are included as features. Notably, short-term trends represented by the 'ema_2' feature (Exponential Moving Average with a span of 2) and the rice index emerge as crucial predictive elements. The 'ema_2' feature reflects short-term changes in food prices, thereby enhancing the model's responsiveness to sudden price fluctuations. The rice index is also a significant feature due to the central role of rice in Indonesian consumption and its substantial contribution to food inflation. These findings highlight the potential of PIHPS data and machine learning techniques in improving food inflation forecasting. However, the study's limitations, including the choice of food categories and the dataset size, indicate the need for further research incorporating diverse data and alternative machine learning models for more comprehensive and accurate predictions.