Stepping up Food Price Prediction using Prophet Model

Abstract

Price fluctuations of strategic food commodities in metropolitan regions like DKI Jakarta present a critical economic challenge, heavily driven by supply chain constraints and localized demand surges during religious festivities. This study addresses the limitation of traditional forecasting models in capturing such non-linear anomalies by implementing the Prophet empirical framework to predict daily food prices across various essential commodities. Utilizing the Cross-Industry Standard Process for Data Mining (CRISP-DM) methodology, the prediction models were rigorously trained and evaluated using Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE) metrics. The empirical results demonstrate that the optimized Prophet model delivers exceptional predictive accuracy for stable commodities, yielding highly accurate scores for garlic (MAPE 5.58%), beef (4.86%), chicken meat (5.73%), chicken eggs (4.55%), sugar (7.01%), and rice (8.61%). Conversely, volatile commodities like curly red chili peppers yielded a moderately accurate performance (MAPE 31.53%). These findings indicate that the structural decomposition approach of the Prophet model offers a highly robust decision-support tool for metropolitan inflation monitoring and market intervention policies.