Combining disparate data sources for improved poverty prediction and mapping

Area of Science:

• Environmental Science
• Data Science
• Socioeconomic Studies

Background:

• Extreme poverty affects over 330 million people in Africa, necessitating precise data for effective policy interventions.
• Existing studies often rely on single data sources, limiting the accuracy and granularity of poverty assessments.
• Big Data offers potential for estimating socioeconomic factors, but integration challenges persist.

Purpose of the Study:

• To develop and validate a computational framework for predicting the Global Multidimensional Poverty Index (MPI) at a fine spatial resolution in Senegal.
• To integrate environmental data and call detail records for improved poverty prediction accuracy.
• To provide policymakers with granular poverty data for designing targeted poverty eradication strategies.

Main Methods:

• Utilized Gaussian Process regression, a Bayesian learning technique, for accurate poverty prediction and uncertainty quantification.
• Employed elastic net regularization for robust model selection and prevention of overfitting.
• Integrated diverse datasets, including environmental factors (food security, economic activity, accessibility) and call data records (CDRs).

Main Results:

• Demonstrated superior prediction accuracy by combining disparate data sources, achieving a Pearson correlation of 0.91.
• Accurately predicted key poverty dimensions—health, education, and standard of living—with Pearson correlations ranging from 0.84 to 0.86.
• Validated all predictions against census-derived deprivation data, confirming the framework's reliability.

Conclusions:

• The proposed framework offers a significant advancement in poverty mapping by leveraging heterogeneous data sources.
• The approach enables frequent generation of high-resolution poverty maps, aiding in the timely design of interventions.
• The diagnostic capabilities of the framework are crucial for assisting policymakers in poverty reduction efforts.