A Comparison of the Holt-Winters Additive and Multiplicative Methods with Grid search Optimization in Forecasting Red Chili Prices in Bengkulu City

Abstract

The Holt-Winters Additive and Multiplicative methods are time series forecasting techniques that consider trend and seasonal patterns. To improve forecasting accuracy, optimization of model parameters, such as alpha, beta, and gamma, is required. This study compares four parameter optimization methods, namely Grid search, Bayesian optimization, Trial and error, and Nelder-mead, through simulation on the generation data and application on weekly red chili price data in Bengkulu City. Red chili was chosen as the object of study because it is a strategic horticultural commodity that has high demand and significant price fluctuations, thus affecting inflation and deflation in Bengkulu City. Simulation results show that the Grid search method consistently produces lower error values and more efficient computation time at iterations 15, 25, 50, and 100. Based on the optimal parameters of Grid search (alpha = 1, beta = 0.1, gamma = 0.1), a Holt-Winters model is obtained that can be used to forecast the price of red chili in Bengkulu City in the first 3 months of 2025. Based on the evaluation of the prediction results, the Additive model provides better performance than the Multiplicative model, with a MAPE value of 11,92% and an RMSE of 7450,52. These findings indicate that the combination of the Holt-Winters Additive method and Grid search optimization is effective in forecasting weekly red chili pepper prices, and can be used as a basis for policy making to control the price of strategic commodities in Bengkulu City

Keywords

• Holt-Winters
• Grid search
• Red Chili
• Statistical optimization
• Box-Jenkins method.