Comparison of Multi Layer Perceptron and Holt Winter Accuracy in Forecasting Suzuki Car Brand Production in Indonesia

Abstract

Car production based on brand holding agents (APM) reports production value to the Indonesian Association of Automotive Industries (Gaikindo) in 2021 of 863,348 units with a percentage range of 11.4% by the Suzuki brand. Based on the public's interest in the need for a car which is the preferred option, the range and selling price are part of the considerations in determining the product of choice as the vehicle owner. This choice is important in activities to meet daily transportation needs. The purpose of this study is to obtain the most effective method and to maintain the number of vehicle unit production, production forecasts are needed according to people's purchasing power patterns using monthly data, especially the Suzuki brand. The research uses the Holt Winters measurement method with two hidden layers on the Multi Layer Perceptron (MLP) measurement method. The findings from this study are comparisons that can be said to be valid and can be an option in predicting data by utilizing methods to predict the amount of production by brand-holding agents. These results can contribute to optimizing the number of production results so that there are no excess or shortage of unit stock. The results showed that forecasting using a Multi Layer Perceptron with two hidden layers produced an accurate value where the lowest value at the Root Mean Square Error (RMSE) was 889.851 and the Mean Absolute Percentage Error (MAPE) was 9.3368.