<![CDATA[Durian production in Dairi Regency, North Sumatra, fluctuates substantially across quarters, creating uncertainty for harvest planning and distribution. This study develops a lightweight decision-support system that integrates an AutoRegressive Integrated Moving Average (ARIMA) forecasting pipeline with a Streamlit web application. The source dataset contains 41,060 agricultural harvest records from 2020–2024, aggregated into 20 quarterly regional observations. The raw series was non-stationary according to the Augmented Dickey–Fuller test (ADF = −0.449, p = 0.901), while first-order differencing produced a stationary series (ADF = −4.120, p = 0.0009). Automated model search selected ARIMA (4,0,1), with AIC = 401.649 and BIC = 407.624. A chronological 80/20 holdout evaluation on the four quarters of 2024 produced an RMSE of 1.16 tons, MAE of 1.15 tons, and MAPE of 4.33%, recalculated from the reported quarter-level forecasts. The Streamlit implementation integrates data management, stationarity diagnosis, automated parameter selection, and forecast visualization. The results indicate that an interpretable ARIMA baseline can provide useful short-horizon regional forecasts when historical data are limited, although the short series requires cautious generalization and further validation.]]>
Copyrights © 2026
