Detecting and Explaining Persistent Road Underdevelopment in Greater Accra Region Using Multi-Temporal Geospatial Data and Machine Learning

Desmond Kemeh , Yuri Ribakov, Israel Klein, Achituv Cohen

Uneven transport infrastructure development remains a persistent challenge in rapidly urbanising cities, where disparities in road conditions shape mobility, accessibility, and socio-economic op-portunities. In Greater Accra, Ghana, rapid urban expansion has produced a road network character-ised by strong spatial inequalities, with many neighbourhood roads remaining unpaved despite sur-rounding development. This study identifies and quantifies persistent road underdevelopment using a reproducible geospatial and machine learning framework integrating OpenStreetMap data with multi-temporal satellite imagery (Landsat-8, Sentinel-1, and Sentinel-2). Road segments were standardised and classified using Random Forest, XGBoost, and LightGBM with spatial cross-validation. Results show that Sentinel-2 combined with LightGBM performed best, achieving a balanced accuracy of about 0.80 and an MCC of about 0.58. Change detection revealed that ap-proximately 11,030.9 km of roads remained persistently unpaved between 2018 and 2024, repre-senting 27.20% of the total road network. Within the urban core, 2,092 km of persistently unpaved roads were identified, compared with 1,593 km of roads upgraded from unpaved to paved. Persistent unpaved roads were spatially clustered, with 14 blind spots identified across 17 districts, covering about 566.9 km². Blind spot areas exhibited significantly higher built-up intensity than non-blind areas, supported by Mann–Whitney (p < 0.001), Welch t-test (p < 0.05), and a large effect size (Cliff’s δ = 0.6211). Moran’s I (0.0803, p = 0.001) confirmed significant spatial clustering. Overall, persistent road underdevelopment is spatially structured and associated with uneven urban devel-opment.