AN INTELLIGENT DEEP LEARNING FRAMEWORK FOR MULTI-CROP PRICE FORECASTING AND AGRICULTURAL SELLING DECISION SUPPORT: A CASE STUDY OF TEN COMMODITIES IN LAMPUNG, INDONESIA

Abstract

Agricultural commodity price volatility presents significant challenges to farmers, traders, and policymakers, particularly in determining appropriate selling prices and managing market uncertainty. This study develops an intelligent deep learning-based Decision Support System (DSS) integrating a Long Short-Term Memory (LSTM) architecture for forecasting ten agricultural commodity prices in Lampung Province, Indonesia. Historical price data were collected from agricultural markets across several districts and processed through data cleaning, outlier handling, normalization, and time-series sequence generation. A multi-output LSTM architecture incorporating stacked recurrent layers, dropout regularization, and fully connected layers was implemented to capture nonlinear temporal dependencies and commodity-specific price fluctuations. Exploratory data analysis, correlation matrices, and temporal visualizations were employed to investigate market characteristics. Experimental results demonstrated promising forecasting performance, achieving an overall Root Mean Squared Error (RMSE) of approximately 0.09 on normalized price data. However, the model exhibited limited sensitivity to abrupt price fluctuations, indicating challenges in predicting highly volatile commodities. The proposed DSS integrates predictive analytics and price visualization to support informed selling decisions, market monitoring, and agricultural risk management. These findings demonstrate the potential of multi-output LSTM forecasting for developing scalable, data-driven agricultural decision support systems in regional commodity markets.