Optimal Short-Term Forecasting Using GA-Based Holt-Winters Method

One of the key issues nowadays in using Holt-Winters Method of forecasting is the appropriate selection of smoothing coefficients. To identify values of smoothing coefficients, an optimization approach is explored that minimizes a forecasting error like Mean Squared Errors (MSE) or Mean Absolute Deviation (MAD). This paper develops a methodology that optimizes forecasting error by determining the optimal smoothing coefficients of the Holt-Winters Method using Genetic Algorithm (GA). This paper focuses on the Mean Square Error (MSE) as an objective value of the optimization problem. The efficiency of the proposed approach was verified using actual test cases based on rice stock commodity in the Philippines. Different variants of the Holt-Winters Method were examined and the result shows that additive seasonal effect was more appropriate for the rice stock data. The proposed approach was compared to other models and the results are promising.