PERBANDINGAN MODEL PROPHET DAN DEKOMPOSISI STL DALAM PERAMALAN FILM BOX OFFICE

Abstract

Box office is the term that used as a movie’s income as well as an indicator of a movie’s success. Forecasting daily box office is important to do because it will be a guidance for producers and distributors in determining a movie’s release date and also as a profit prediction. Daily box office has a large amount of historical data and strong seasonality. Prophet model and seasonal-trend decomposition using Loess (or simply STL decomposition) are some of the forecasting methods that are capable for forecasting daily data with large frequency and strong seasonalities. The goal in this research is to observe the comparison between prophet model and STL decompostion and also to demonstrate each method’s computation to further elaborate on each method’s forecasting performance. The result in this research shows that a modification of prophet model with the addition of holidays as a parameter has the lowest error using root mean square error (RMSE) evaluation with a score of 7.515.225 and using mean absolute percentage error (MAPE) evaluation with a score of 33.62%.