COMPARATIVE ANALYSIS OF THE GOODNESS OF GSTAR IN SPATIAL DEPENDENCY WITH DISAGGREGATION METHOD

• Main Author: Raeva Maulana, Fauzan
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/57736
• Institution: Institut Teknologi Bandung

The GSTAR model or commonly known as generalized space-time autoregressive is a space-time model that not only calculates in terms of observation time, but also calculates in terms of the influence between locations. The influence between locations is identified by the spatial weight matrix variable. Each spatial effect is compared with each other in order to obtain the best GSTAR model that describes the data at the location under review. The GSTAR model is disaggregated by using the disaggregation method. The disaggregation method is used to predict data at lower levels. In this final project, the increase in COVID-19 cases in each sub-district in the East Jakarta Administrative City is modeled with an observation time from September 2, 2020 to January 28, 2021 to obtain results from additional cases of COVID-19 at the urban villages level. The GSTAR model uses three kinds of weight matrices, namely uniform, Euclidean distance, and normalized cross-correlation. The GSTAR model uses GSTAR(1;1) which means time order 1 and spatial order 1. Each GSTAR model is disaggregated by selecting a dummy variable that is positively correlated with COVID-19 case data. The results of the GSTAR model with the disaggregation method are compared with the actual data. As a result, by analyzing MAE, MSE, and MAPE, it is found that the GSTAR(1;1) model for each weight matrix is not much different. This is due to the very random movement of population between residents in East Jakarta.