SISTEM PREDIKSI LONJAKAN HARGA PANGAN BERBASIS RANDOM FOREST UNTUK EARLY WARNING SYSTEM

Abstract

Food prices are essential indicators of economic stability and public welfare. Uncontrolled fluctuations, particularly sudden spikes, lower purchasing power and drive inflation. Currently, food price information systems are largely descriptive and lack predictive capabilities for sudden anomalies. To address this gap, this research develops a web-based Early Warning System (EWS) featuring digital integration and automation to proactively detect price spikes. The study utilizes daily secondary time-series data from the Strategic Food Price Information Center (PIHPS) via Kaggle, spanning from 2022 to 2026. Feature engineering, including 7-day moving averages and percentage changes, was applied to enhance the Random Forest classification algorithm. To handle the extreme 95:5 data imbalance, a class-weight balancing technique was employed during modeling. The empirical findings demonstrate that the model achieved an overall accuracy of 74% and a recall of 56% for the minority 'spike' class, proving its capability to capture more than half of the actual market crises. Furthermore, feature importance analysis revealed that the 7-day moving average was the most dominant predictor, contributing 26.22% to the model's decisions, which indicates the system effectively recognizes historical market volatility rather than nominal values. This prediction engine was successfully integrated into an interactive Laravel-based dashboard equipped with automated alert notifications. However, the system's high sensitivity resulted in a low precision of 9%, generating frequent false positive alerts that require further architectural refinement to mitigate alert fatigue in future studies.