All results are based on the INDICATORS derived from DATA SOURCES by using machine learning methods to link extracted information with national statistics data.
Predicted values come from machine learning model and estimated values from the national statistics office.
Check differences above or below main diagonal – indicators unveil their hidden population trends.
Municipalities ranked by predicted depopulation risk, from low to high, can help in prioritizing interventions.
Municipalities denoted with white nodes have positive population change rate, while denoted in black have negative.
Importances of indicators are derived from machine learning (random forest) predictive model learned on fusion of all data. With recursive feature elimination the set of indicators is reduced to 30 the most important.
Indicators derived from all data sources jointly contribute to the prediction of population change rate. Highly important are closeness centralities from connectivity, betweenness centrality from mobility, industrial, commercial and urban unitis from corine, residential roads density from OpenStreetMaps,internet activity in the night hours.
Importances of indicators are derived from machine learning (random forest) predictive model learned on activity data.
Information on type of day, activity during Covid19 lockdown and roaming increase the importance of activity indicator.
Importances of indicators are derived from machine learning (random forest) predictive model learned on connectivity data.
Closeness centrality is an important indicator implying that municipalities with high values of this indicator have better access to information and more influence on other municipalities. These municipalities have better population trends.
Importances of indicators are derived from machine learning (random forest) predictive model learned on mobility data.
Indicators related to clustering and betweenness centrality in mobility flows have high importance. Municipalities with only pronounced local mobility flows (large clustering) show signs of isolation and are more severely affected by depopulation. Contrary, municipalities with high betweenness centrality representing mobility hubs and bridges between municipalities have high in-migrations.
Importances of indicators are derived from machine learning (random forest) predictive model learned on Corine land use data.
The most important indicators represent percentage of the discontinuous urban fabric and industrial/commercial units land use class.
Importances of indicators are derived from machine learning (random forest) predictive model learned on OpenStreetMap data (point of interest and roads data).
Density of all roads and in particular residential roads, as well as number of all points of interest have high importance for the depopulation context.
Summary plot combines indicator importance with indicator effects. Each point on the summary plot is municipality. The position on the y-axis is determined by the indicator and on the x-axis by the SHAP (Shapley value) that indicates its effect. The color represents the value of the indicator (feature) from low to high.
Blue dots denote high value of the indicators, while red denote small. Positive SHAP values indicate better population trend, while negative warns on depopulation.