Yan Watequlis Syaifudin, Alvina Marcy Syakirah Permata, Sujarwo, Triana Fatmawati, Pramana Yoga Saputra, Indrazno Siradjuddin, Yudi Widhiyasana, Yuri Ariyanto, Abiyasa Putra Prasetya
Agricultural commodities are essential for food security and livelihoods, supporting millions of small-holder farmers and traders. Some commodities are highly susceptible to price volatility due to seasonal production, weather conditions, and market demand. These fluctuations significantly impact inflation and household purchasing power, which makes accurate price prediction crucial for effective decision making across the agricultural supply chain. Recent advances in deep learning have positioned LSTM networks as a powerful solution, outperforming traditional models by capturing nonlinear patterns and long-term dependencies in time-series data. This paper presents an experimental study on end-to-end forecasting framework that integrates real-time data collection from an official website, automated preprocessing, and a two-layer LSTM model to predict prices up to 30 days ahead, with evaluation metrics showing high accuracy for stable commodities like garlic while revealing challenges in forecasting sudden spikes in more volatile items such as chili peppers and red chilies. The results confirm LSTM's effectiveness over other naïve models, demonstrating its potential to support farmers, traders, and policymakers, especially when applied within a system that bridges data acquisition, modeling, and visualization. © 2025 IEEE.
State Polytechnic of Malang, Dept. of Information Technology, Malang, Indonesia; Brawijaya University, Dept. of Socio-economics Agriculture, Malang, Indonesia; State Polytechnic of Malang, Dept. of Electrical Engineering, Malang, Indonesia; State Polytechnic of Bandung, Dept. of Informatics and Computer Engineering, Bandung, Indonesia
