Résumé: |
Geodemographic data mining is a practice of classifying and characterizing the population
living in a specific area. As geodemographic problems, such as population dynamics, are
rather complex, using a quantitative approach based on machine learning can enable
the extraction of some insightful information. In this study, we present a methodology
to characterize patterns of population dynamics in Switzerland from 2000 to 2020. This
starts with an exploratory analysis of some geodemographic variables and some clustering
algorithms to determine the most accurate procedure. During this phase, a descriptive
analysis helped to deal with the question of feature selection. Then, we compared two
unsupervised algorithms, that enable the understanding of this high-dimensional space
thanks to a clustering analysis. Those algorithms are different a k-means clustering and a
Self-organizing neural network. The result of this exploratory phase has led to the choice
of a Self-organizing map with around ten input variables to describe population dynamics
in Switzerland in recent times. Those inputs, aggregated at the municipality level, come
from the Swiss national census and contain both socioeconomic information on the population
and features about land use. The results using this method looks similar to the
typology of urban areas described in other research. They also illustrate the evolution of
the sprawling cities, the homogenization of the Swiss plateau, and the constraints of the
Alps over the dynamic of the population. |