DYNAMIC GRAPH REPRESENTATION LEARNING FOR FLIGHT DATA AS A FAST ECONOMIC INDICATOR

Abstract

Future economic disruptions, such as natural disasters or pandemics, can impact a country's economy rapidly. However, current GDP-based economic indicators are slow to compile, taking months to release. This delay makes GDP growth unsuitable for timely policy decisions. A faster economic indicator is needed, one that correlates with official GDP, uses open data, and maintains historical consistency for predictive purposes. Big data, including shipping data, population mobility, and sentiment analysis, is being used to predict economic growth in near real- time. Air travel, a significant mode of transport in trade and tourism, has shown strong correlations with economic growth. Research indicates that the presence of airports and increased flight frequencies contribute to local economic activity. Flight data, which forms a global network between airports, can be modeled as a graph, where nodes represent airports and edges represent flight connections. Using graph analysis methods, these connections can be explored, and representation learning can automatically generate vector embeddings for nodes, simplifying
the analysis of large and complex networks. These embeddings can be further used in machine learning models for economic predictions.