GENERALIZED SPACE TIME AUTOREGRESSIVE (GSTAR) MODELING WITH MINIMUM SPANNING TREE (MST) SPATIAL WEIGHT MATRICES IN TRAFFIC FLOW

• Main Author: Yolanda, Feby
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/72969
• Institution: Institut Teknologi Bandung

Space-time model is a model that is not only influenced by observations in previous times but also by observations in nearby locations. One of the most frequently used space-time models is GSTAR. This model is a development from AR, VAR, and STAR model. What makes GSTAR or other space-time models different from time-series model is its spatial weight matrices. Spatial weight matrices which developed by considering nearest locations are used to describe the relationship of one location with other locations. Spatial weight matrices can also be built with Minimum Spanning Tree (MST) to emphasize probability relationship between locations. One example of space-time data is traffic flow because it is influenced not only by past data but also by data on the surrounding road. Therefore, this Undergraduate Thesis will analyze GSTAR model with MST spatial weight matrices in traffic flow. The aim of the modeling is to predict the road occupancy on certain roads in the future that can be used for quantification of the risk of a road. The modeling will be done by forming the MST spatial weight matrices and conventional spatial weight matrices. Next step is ????????????????????(1;1) modelling which includes parameter estimation using least square method until prediction. By using AIC dan MAPE criterion, it can be concluded that ????????????????????(1;1) model with conventional spatial weight matrices is better than model with MST spatial weight matrices.