TY - GEN
T1 - MGSTARX Model for Forecasting Space-Time Data with Metric Exogenous Variable
AU - Nahdliyah, Nurun
AU - Setiawan,
AU - Rahayu, Santi Puteri
N1 - Publisher Copyright:
© 2023 American Institute of Physics Inc.. All rights reserved.
PY - 2023/1/27
Y1 - 2023/1/27
N2 - The Multivariate Generalized Space-Time Autoregressive (MGSTAR) model is a model that is used to forecast space-time data with several variables in several locations. MGSTAR model has been developed into hybrid MGSTAR-ANN model for non-linear cases and hybrid MGSTARX-RNN model for cases with calendar variation effect. The calendar variation effect is a non-metric exogenous variable. This study aims to propose the MGSTARX model for cases with metric exogenous variables. There are two steps of MGSTARX modeling. The first step is modeling the data that involve exogenous variables using two approaches, i.e., Time Series Regression (TSR) and Transfer Function (TF). Then, the residuals from the first step are modeled using MGSTAR. This study focused on a simulation study to evaluate the goodness of the MGSTARX model. The result shows that the MGSTARX model with the transfer function approach is more accurate and has the smallest RMSE for forecasting the data than the other model. In general, the MGSTARX model with an exogenous variable can improve the accuracy of forecasting. This result is in line with the results of the M5 Accuracy Competition. Further study is needed to expand the MGSTARX model for higher dimensions data and other patterns such as seasonal and non-linear.
AB - The Multivariate Generalized Space-Time Autoregressive (MGSTAR) model is a model that is used to forecast space-time data with several variables in several locations. MGSTAR model has been developed into hybrid MGSTAR-ANN model for non-linear cases and hybrid MGSTARX-RNN model for cases with calendar variation effect. The calendar variation effect is a non-metric exogenous variable. This study aims to propose the MGSTARX model for cases with metric exogenous variables. There are two steps of MGSTARX modeling. The first step is modeling the data that involve exogenous variables using two approaches, i.e., Time Series Regression (TSR) and Transfer Function (TF). Then, the residuals from the first step are modeled using MGSTAR. This study focused on a simulation study to evaluate the goodness of the MGSTARX model. The result shows that the MGSTARX model with the transfer function approach is more accurate and has the smallest RMSE for forecasting the data than the other model. In general, the MGSTARX model with an exogenous variable can improve the accuracy of forecasting. This result is in line with the results of the M5 Accuracy Competition. Further study is needed to expand the MGSTARX model for higher dimensions data and other patterns such as seasonal and non-linear.
UR - https://www.scopus.com/pages/publications/85147271941
U2 - 10.1063/5.0106032
DO - 10.1063/5.0106032
M3 - Conference contribution
AN - SCOPUS:85147271941
T3 - AIP Conference Proceedings
BT - 3rd International Conference on Science, Mathematics, Environment, and Education
A2 - Indriyanti, Nurma Yunita
A2 - Sari, Meida Wulan
PB - American Institute of Physics Inc.
T2 - 3rd International Conference on Science, Mathematics, Environment, and Education: Flexibility in Research and Innovation on Science, Mathematics, Environment, and Education for Sustainable Development, ICoSMEE 2021
Y2 - 27 July 2021 through 28 July 2021
ER -