Predicting poverty and wealth from Mobile phone metadata
Research paper by prof. Joshua Blumenstock
- Mobile phone use reflects the structure of individual’s social network, patterns of travel and location choice and histories of consumption and expenditure
- Survey on asset ownership, housing characteristics and other welfare indicators.
- Constructed a composite wealth index
- Mobile phone data is used to predict the wealth index calculated from survey data
- Features constructed are:-
- Total volume
- Intensity
- Timing
- Direction of communication etc
- Structure of the individual’s contact network
- Patterns of mobility based on geospatial markers
- Elastic Net regularization was used in modelling
- Geospatial markers in the phone data enabled to study the geographic distribution of subscriber of wealth at an extremely fine degree of spatial granularity
- There was a strong correlation between the mobile metadata predictions and the DHS survey data at district and village levels. Correlations persisted even for comparing clusters within urban and rural areas
- This approach can be used to predict other metrics as well. Rates of district electrification estimated from phone records are comparable to those reported in the DHS survey
