Implementasi ANFIS Untuk Forecasting Penjualan Sembako Pada CV XYZ

Abstract

This study compares the performance of three fuzzy membership functions—Gaussmf, Gbellmf, and Trimf—in forecasting basic goods sales. The evaluation was conducted by measuring the Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE) on both training and testing data across various window sizes. The evaluation results indicate that the Trimf membership function achieved the best performance. For a window size of 4, Trimf yielded a testing RMSE of 1.77 and a testing MAPE of 8.23%, outperforming Gaussmf (RMSE 2.82, MAPE 12.57%) and Gbellmf (RMSE 4.64, MAPE 17.54%). Meanwhile, Gaussmf and Gbellmf exhibited weaker performance on testing data, particularly at larger window sizes. These findings suggest that the appropriate selection of fuzzy membership functions can significantly enhance prediction accuracy. Future research could explore combinations of membership functions or other parameters that may further improve forecasting performance.