Forecasting Ginger Harvest Yields: A Comparative Study of Double Exponential Smoothing and Long Short-Term Memory Models

Ginger, a vital herbal commodity, experiences low yield rates, necessitating intensive cultivation and rigorous evaluation by farmers to ensure financial viability and alignment with market demands.This study was conducted to devise a harvest forecasting system that supports decision-making through minimal error rates by comparing double exponential smoothing (DES) and long short-term memory (LSTM) forecasting methods.The efficacy of these methods was assessed through a series of trials, analyzing data collected from 2015 to 2019, comprising 250 datasets.The evaluation focused on two primary metrics: the Mean Absolute Percentage Error (MAPE) and the Root MSE (RMSE), to determine the precision of forecast models.It was observed that the LSTM model outperformed the DES method, yielding a MAPE of 38.99% and an RMSE of 1244.85432, in contrast to the DES method which resulted in a MAPE of 43.49% and an RMSE of 12997.34261, at an alpha level of 0.4 and an optimal beta of 0.1.Given these findings, the LSTM model is recommended for the forecast of ginger yields due to its superior accuracy and lower standard error compared to the DES method.This comparative analysis underscores the importance of selecting appropriate forecasting models to enhance agricultural planning and productivity, particularly in crops with fluctuating yields such as ginger.