Using Drones and Machine Learning to Build Kananga's First Modern Property Tax Roll, D.R. Congo

Since June 2023, researchers from the Local Government Revenue Initiative (LoGRI) and UC Berkeley have worked with Kasaï-Central province's tax authority, the Direction Générale des Recettes du Kasaï-Central (DGRKAC), to rebuild the property tax roll for Kananga, a city of roughly one million people that had never had a systematic one. The project used drone imagery and AI-based rooftop detection to estimate building footprints and floor counts across the city, combined with a field survey of property characteristics, then trained machine-learning models on 1,654 properties independently valued in person by government land surveyors to predict values for the remaining stock — a computer-assisted mass appraisal (CAMA) approach designed for a context with very limited administrative data.

Published as ICTD Working Paper 176 (Institute of Development Studies, November 2023), the results show the best model, trained on administrative property characteristics, reached a cross-validated R² of 60%, with 22% of predicted values landing within 20% of the surveyors' benchmark; a computer-vision model using only property photographs performed markedly worse, at 9% within the same band — an honest result the researchers describe as evidence that simple machine-learning methods can build a usable tax roll even where property information is scarce, not as a finished, highly accurate valuation tool. The drone survey identified 106,111 assessable buildings citywide. In April 2025, LoGRI ran a technical handover workshop transferring operation of the resulting Moptax digital tax-management platform — covering valuation, billing, mobile-money payment and appeals — to DGRKAC staff directly, moving the system from externally supported pilot to local government ownership for its first full year of independent operation.

Read the full analysis: https://www.ictd.ac/publication/using-machine-learning-to-create-property-tax-roll-evidence-from-kananga/