GENERALIZED SPACE TIME AUTOREGRESSIVE MODELING WITH MINIMUM SPANNING TREE SPATIAL WEIGHT MATRICES (CASE STUDY ON MONTHLY PASSENGERS NUMBERS AT 17 AIRPORT IN AUSTRALIA)

• Main Author: Cantika Gumilar, Yasmin
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/84132
• Institution: Institut Teknologi Bandung

The airline industry is a rapidly growing transportation sector that plays a crucial role in the global economy by creating jobs and supporting the tourism and manufacturing sectors. During the COVID-19 pandemic period, the airline industry was severely impacted by a significant decrease in the number of flights. Air travel demands may fluctuate, increasing during holiday seasons and sharp declines during global crises. The ability to analyze and predict air travel movements is essential for maintaining an operational efficiency and an infrastructure readiness. In this Final Project, to model the number of passengers at an airport, it is assumed that the number of passengers at that airport dependent upon the number of passengers at that airport in previous periods and those at nearby airports. The GSTAR model with an inverse distance weight matrix and a minimum spanning tree weight matrix is applied to model the monthly number of passengers at 17 airports in Australia, namely airports in Adelaide, Brisbane, Cairns, Canberra, Gold Coast, Hamilton Island, Hobart, Karratha, Launceston, Mackay, Melbourne, Perth, Rockhampton, Sunshine Coast, Sydney, and Townsville. The modeling process begins with data preparation in a format suitable for data analysis; construction of the inverse distance weight matrix and the minimum spanning tree; model identification; parameters estimation using the least squares method; and diagnostic testing. The fitted model is then used to predict the number of passengers at a specific airport at a particular time. For the case study data, using the inverse distance weight matrix, the best GSTAR model obtained is GSTAR(3;4,2,4), with the estimated parameters given in Chapter III of this final project. When using the minimum spanning tree weight matrix, the best model obtained is GSTAR(3;4,2,3), with estimated parameters also given in Chapter III of this final project. Based on the AIC, RMSE, and MAPE criteria, as well as an analysis on the residual patterns and the number of the estimated parameters, it could be concluded that the GSTAR(3;4,2,3) model with the minimum spanning tree matrix is better than the GSTAR(3;4,2,4) model derived using the inverse distance weight matrix.