Perbandingan Metode ARIMA(Autoregressive Integrated Moving Average) dan GAMLSS(Generalized Additive Model for Location,Scale,and Shape) Dalam Memprediksi Kunjungan Situs Website (Studi Kasus: Website Laboratorium Statistika Universitas Jember)

Abstract

Zero inflation is a condition where data with a value of zero is recorded more than expected, which often occurs due to events that have no activity or movement and are ultimately ignored. Data with zero inflation characteristics is sometimes difficult to predict. This research aims to find a suitable method for predicting time series, especially for time series data with excess zero inflation. The methods used in this research are ARIMA and GAMLSS. ARIMA is a classic method in time series analysis that relies on autoregressive, differencing and moving average components to capture patterns in data. Meanwhile, GAMLSS is a more flexible method that allows modeling more complex data distribution parameters. In this research, both methods are applied to historical zero inflated data and their performance is evaluated using metrics such as Root Mean Square Error (RMSE) and using the Aikaike Information Criterion and Schwarz Bayessian Criterion(SBC) criteria. The results show that the best model for data with excess zero inflation is GAMLSS with a special Zero Inflated Negative Binomial (ZINBI) distribution.