Mobility diversity and socio-economic development Luca Pappalardo www.lucapappalardo.com @lucpappalard University of Pisa ISTI-CNR
Surveys have limits They do not scale to small territories and they are not dynamic They do not adapt to Pareto distributions: We cannote use samples 50 people are as wealthy as 3.5 billion people Young people do not respond to surveys
How to measure well-being in an automatic way (without surveys)
Big Data The GPS tracks of car travels originating in Pisa (red) vs. the GPS tracks of car travels originating in Florence (green) (viz by KDD LAB)
Advantages of Big Data They capture the complexity of social systems in their entirety They allow for the observation of complex phenomena, like diversity, resilience and equality
An analytical framework
Data format timestamp coords caller callee 04/01 23:45:00 132.56, 23.64 145323 452300 04/02 06:02:00 143.28, 54.22 5602 …
Data format timestamp coords caller callee 04/01 23:45:00 132.56, 23.64 145323 452300 04/02 06:02:00 143.28, 54.22 5602 …
Orange dataset 20M users 5.7G calls 87K towers 45days
Mobility volume: characteristic distance traveled by an individual Mobility diversity predictability of an individual’s movements
differently diversified
Deprivation index Per capita income - Home location is the most frequent during nighttime (8 pm - 3 am) - Every individual’s home is assigned to its municipality - Measures are aggregated at municipality level - Economic measures are collected Deprivation index Per capita income
Mobility diversity and well-being We investigate the correlations between the aggregated mobility measures and the four external socio-economic indicators. We see from the figures that a clear tendency emerges, in this case for mobility entropy: the higher the mobility entropy the lower the deprivation index.
Data NM1 NM2 $$ $$ $$ $$ $$ $$ $$$ $$ $ Paris Marseilles Lyon Paris
NM1 NM2
NM1 NM2
Predicting well-being Multivariate Regression predicting the exact value R2 = 0.42 (deprivation) R2 = 0.25 (income) Classification: predicting class of well-being (low, medium, high) acc = 0.61 (deprivation) acc = 0.54 (income)
Diversity matters mobility diversity mobility diversity
International Journal of Data Science and Analytics The purpose of the study I am gonna present you today is to understand the relationships between human mobility and the socio-economic development of a territory. In order to do that we performed a data-driven study in France exploiting the access to mobile phone data provided by the Orange Telecom provider. International Journal of Data Science and Analytics https://link.springer.com/article/10.1007/s41060-016-0013-2
Take home message(s) Big Data allow for the observation of complex phenomena (e.g. human mobility) Mobility diversity is linked to aspects of well-being A city’s well-being can be predicted from measures extracted from Big Data
Grazie! Luca Pappalardo University of Pisa ISTI-CNR www.lucapappalardo.com www.sobigdata.eu @lucpappalard University of Pisa ISTI-CNR