MODEL RADIAL BASIS FUNCTION NETWORK - BASED STATE - DEPENDENT AUTOREGRESSIVE UNTUK PERAMALAN DATA RUNTUN WAKTU MUSIMAN DAN TREN ( RADIAL BASIS FUNCTION NETWORK - BASED STATE - DEPENDENT AUTOREGRESSIVE MODEL FOR SEASONAL AND TREND TIME SERIES FORECASTING )

SEPTIANA, DIAN, Herni Utami

2016 | Disertasi | FMIPA

There are many time series data possess seasonal and trend patterns. For instance, data in climate and economic sectors. In this case, a proper forecasts will provide much benefit for other real life sectors. The studies on the concern of seasonal and trend time series data lead to further discussion. Previous investigations revealed that this kinds of data are linear, but others found that seasonal and trend pattern in time series data are non linear. Hence, an appropriate forecasting method is necessary to obtain an ideal result. This research will mainly described about Radial Basis Function - Based State - Dependent Autoregressive (RBF-AR) model, which theoretically is capable to capture non linear factor in a data and will give a good accuracy. In this model, autoregressive parameter is compiled by using Radial Basis Function (RBF) network. Overall, these models are optimized with Structured Nonlinear Parameter Optimization Method (SNPOM) algorithm which divides the parameters into 2 parts, linear and nonlinear. Experiment result shows that forecasting for some step forward need a proper amount of data, which is not too long, or not too short. In this research, proper precipitation data is 10 years data. Precipitation forecast data with 12 seasonal periodand milk production data whose seasonal period is 12 and trend were done by comparing 3 forecasting model, viz. RBF-AR, RBF, and SARIMA. For precipitation data, RBF-AR bring about better RMSE value than RBF and SARIMA. Whereas for milk production data, SARIMA is better than two others

Kata Kunci : Forecasting, Time Series, Seasonal, Trend, RBF-AR, RBF, SARIMA