Poverty prediction using E-commerce dataset and filter-based feature selection approach.

Area of Science:

• Socioeconomics
• Data Science
• Computational Social Science

Background:

• Poverty assessment traditionally relies on time-consuming surveys and censuses.
• Governments require rapid insights into socio-economic conditions for effective development planning.
• Existing methods for poverty data collection are resource-intensive and slow.

Purpose of the Study:

• To develop a faster proxy for poverty level assessment using e-commerce data and machine learning.
• To investigate the efficacy of various machine learning algorithms in predicting poverty rates.
• To propose a novel approach combining feature selection with predictive modeling for poverty mapping.

Main Methods:

• Utilized a high-dimensional e-commerce dataset for poverty prediction.
• Employed statistical-based feature selection algorithms (e.g., f-score) to identify relevant predictors.
• Compared three machine learning models: support vector regression, linear regression, and k-nearest neighbor.

Main Results:

• The combination of f-score feature selection and support vector regression demonstrated superior performance in predicting poverty rates.
• E-commerce data, when processed with appropriate feature selection, proved effective for poverty level estimation.
• The proposed methodology offers a significant speed advantage over conventional survey-based approaches.

Conclusions:

• E-commerce data and machine learning algorithms present a viable and efficient proxy for poverty level prediction.
• The integration of feature selection enhances the accuracy and efficiency of machine learning models for socio-economic analysis.
• This approach can provide timely information for policymakers to inform area development strategies.