Using machine learning to value property in Kigali, Rwanda

Kurzfassung

This paper develops two property valuation models for Kigali, Rwanda, and tests them on a unique dataset combining remote sensing data for buildings in Kigali, with sales transaction data for 2015. This paper credits and builds on a similar paper by Deininger et al (2018) but also covers both the built up area of Kigali and the whole of Kigali Province, it addresses temporal prediction issues beyond 2015, it employs an expanded set of variables, and it uses machine learning techniques to employ Maximum Relevance Minimum Redundancy to select the model that best predicts property price data using Ordinary Least Squares. The model in this paper is intended as a prototype of a Computer Assisted Mass Appraisal system for Kigali that could be used to calculate the revenue potential of a new property tax introduced in Rwanda in 2019, and to help detect under-declaration of property values for tax purposes.