FORECASTING THAI DURIAN EXPORTS USING A HYBRID TIME SERIES SARIMA-SVR APPROACH

Thailand is a prominent frontrunner in Southeast Asia in terms of its production and exportation of durian. Each year, the production of durian is a significant source of income for farmers. The consumption of durian has continued to increase throughout the years, both within the country and internationally. Consequently, cultivators have increased their durian cultivation to meet the expanding customer demand. However, there is a chance of excessive production, which has the ability to disrupt future market trends. Precise forecasts of durian export volumes are crucial for planning production strategies. This work aims to develop a novel SARIMA-SVR model for forecasting future export quantities of Thai durian, thereby fulfilling this need. The approach involves using the empirical mode decomposition (EMD) technique to reduce data fluctuations. The work conducts an analysis using the SARIMA model and calculates Gaussian white noise, which we then use as an input variable for SVR modeling. The study creates and evaluates four hybrid models, specifically SARIMASVR1, SARIMASVR2, SARIMASVR3, and SARIMASVR4. The study’s findings suggest that SARIMASVR4 demonstrates superior accuracy compared to other models. Furthermore, they show the effectiveness of improving forecast accuracy by utilizing data filtering techniques such as EMD and incorporating Gaussian white noise.