An Optimized Machine Learning Model for Poverty Detection Using a Multidimensional Data-Driven Approach
Abstract / Description
Accurate and dynamic poverty detection remains a major challenge where conventional approaches rely on static classifiers and unimodal data, resulting in limited adaptability and reduced predictive accuracy. This research introduces the Multidimensional Data-Driven Approach (MDDA), an optimized machine learning framework that combines hybrid GA-PSO-GWO optimization, graph-based multimodal data fusion, and ethical AI mechanisms incorporating differential privacy and bias mitigation. The framework integrates satellite imagery, mobile metadata, and household survey data to support real-time poverty detection. Experimental evaluation demonstrated 93.2% classification accuracy, an AUC-ROC of 0.946, improved fairness across vulnerable populations, resilience to concept drift, and significant potential for improving the efficiency of public resource allocation. The proposed MDDA provides a scalable and policy-oriented framework for supporting Sustainable Development Goal 1 through accurate, adaptive, and evidence-based poverty monitoring.
Keywords
Suggested Citation
Interested in Research Collaboration?
Partner with Global Signature Consultancy for research, surveys, data analytics, institutional studies and evidence-based reporting.
Contact Me