Modeling of a Multidimensional Data-Driven Approach (MDDA) for Optimized ML Model in Poverty Detection
Abdulrehman Mohamed, Fullgence Mwachoo Mwakondo, Kevin Tole, Mvurya Mgala · The International Journal of Engineering and Science · 2025
Poverty detection remains a critical challenge in socio-economic development, necessitating innovative, scalable, and efficient methodologies for accurate assessment and intervention. Traditional poverty assessment techniques, such as household surveys and economic censuses, suffer from limited scalability, delayed updates, and inherent biases, reducing their effectiveness in dynamic socio-economic landscapes. Advances in Machine Learning (ML) and big data analytics offer promising alternatives by integrating multimodal data sources, including geospatial information, mobile network metadata, financial indicators, and social media analytics. However, existing ML-based poverty detection models face challenges in real-time adaptability, bias mitigation, computational efficiency, and scalability. This study introduces the Multidimensional Data-Driven Approach (MDDA), an optimized ML framework that integrates multimodal data fusion, fairness-aware ML techniques, and hyperparameter optimization to improve poverty classification accuracy. The MDDA methodology follows five key phases: synthetic data generation and preprocessing, feature engineering and selection, ML model development, bias mitigation, and performance evaluation. The approach is tested on a synthetic dataset of 100,000 records, simulating socio-economic indicators across diverse geographic and economic contexts. Performance evaluation metrics include classification accuracy, fairness measures (Demographic Parity, Equalized Odds), computational efficiency, and real-time adaptability. Experimental results confirm that MDDA achieves a classification accuracy of 91.2%, reduces bias by 15-20%, and improves computational efficiency by 30% compared to baseline ML models. Additionally, MDDA is compared against established ML approaches such as CRISP-DM, SCRUB, KDD, TDSP, SEMMA, and KID, demonstrating superior performance in real-time adaptability, bias mitigation, multimodal data integration, and scalability. These findings highlight MDDA as a real-time, unbiased, and scalable solution for poverty detection, with direct implications for policymaking, economic planning, and humanitarian aid distribution. The study underscores the transformative potential of AI-driven poverty classification, bridging the gap between fairness, efficiency, and real-time adaptability in socio-economic analysis