CONSTRUCTION OF MULTIVARIATE GENERALIZED SPACE TIME AUTOREGRESSIVE (MULTIVARIATE GSTAR) MODEL WITH MODIFIED WEIGHT MATRIX

• Main Author: Volisa, Metra
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/61882
• Institution: Institut Teknologi Bandung

Since COVID-19 first appeared in China, the virus has grown so fast and spread around the world in a matter of months that the WHO (World Health Organization) has designated the COVID-19 virus as a global pandemic. The unrest experienced by all countries requires their governments to take policies to suppress the transmission of the virus from one location to another. A space-time model can be designed to predict COVID-19 cases in various locations with various observation variables. The Multivariate GSTAR model was constructed to predict several variables and locations simultaneously with a sequence of observations based on time. The weight matrix was built using the inverse of distance and correlation between cases at each observation location which was then modified. The model that has been obtained is then estimated using the least-square method. The process stationary examination used residual test, parameter matrix eigenvalue approach, and inverse autocovariance matrix approach. The model is applied to predict COVID-19 infected, death, and recovery cases for all provinces on the island of Sumatra. The results showed that the Multivariate GSTAR (1;1) model was very well applied in predicting death cases in the province of Bangka Belitung Islands and Bengkulu.