A computational framework for analyzing urban data usage patterns in African cities: a 25-year data-driven review using natural language processing and machine learning

Abstract

This study provides a scoping review of urban data usage in African cities over the past 25 years, leveraging advanced Natural Language Processing − NLP, Machine Learning − ML, and hybrid approaches to classify urban data. The fine-tuned classification framework demonstrated robust performance, achieving an accuracy of 91.76 % in classifying relevant abstracts, with calibrated confidence scores ensuring reliable and evidence-aligned predictions. Topic modeling analysis is used combined with a personalized dictionary to extract five urban data typologies: Spatial, Digital, Commercial, Public Sector, and Sensor Data. These typologies are mapped to 15 urban contexts, revealing significant regional disparities that offer deeper insights into local practices, successes, and deficiencies. The model’s ability to capture well-defined contexts, such as Urban Settlement & Housing, highlights its strength, while lower confidence in overlapping themes, such as Urban Environment & Climate, underscores the complexity of these categories. Rigorous validation, including stratified k-fold cross-validation and stability testing of topic modeling parameters, ensures the replicability and generalizability of the framework. To move beyond descriptive comparisons, we conducted a chi-square analysis, which revealed a statistically significant but modest association between geographic regions and urban research themes across Africa (χ 2 = 219.88, df = 36, p < 0.001; Cramér’s V = 0.120), confirming that observed regional disparities reflect genuine differences in research priorities rather than random variation. The analysis reveals three continental patterns in urban research: Environment & Climate dominates overall but is unevenly distributed, with strong concentration in Southern and North Africa, while Health & Sanitation shows a clear East/West versus North/South divide. Infrastructure research exhibits the greatest regional inequality, with relative specialization in Southern Africa and under-representation in East and West Africa, likely to reflect differences in research capacity, funding, and development trajectories. These findings not only provide actionable insights into regional urban research priorities but also establish a replicable methodology for systematically analyzing urban data in diverse and resource-constrained settings.