The Methodological Expansion for Global City Analysis: From Principal Component Analysis and k-Means Clustering to Self-Organizing Map

Research for globalization and global cities has increased the variables on the catalog to explain globalization and global cities. This expansion of global city data requires a more efficient method which can simplify the process and understand analysis intuitively. The purpose of this study is to test a new SOM (self-organizing map) as an addition to global cities analytic methods comparing to the traditional method based on principal component analysis and k-means clustering. Most available 15 attributes for 75 global cities are collected for the research purpose. The principal component analysis found 4 factors, and global cities are classified into 5 types by k-means clustering. The self-organizing map analyzed the same data and visualized the clusters on U-matrix (unified distance matrix) and the patterns of the input attributes on component planes. Compared to the traditional process of data reduction, the SOM effectively visualize common attributes on component planes preserving the topological properties so that it is easy to analyze the relationship between attributes and cities. However, the outputs of the case classification are different from the outputs of the traditional methods. Most importantly, the SOM provides data-driven research with fewer decisions made by analysts than the traditional process.