Fine-scale mapping of residential land price using machine-learning: An experimental study in the city dominated by informal land markets.

Abstract

Context and backgound Fine-scale mapping of residential land price (RLP) is essential to the understanding of residential land market dynamics and improving urban planning. However, such cartographic resources and experimental studies to map RLP at fine-scale in Sub-Saharan African cities are limited as a result of informal land market dominance in shaping the growth and expansion of most of the cities in the region. Goal and Objectives: The study seeks to establish an optimized ensemble machine-learning method for mapping RLP at grid-level in Dar-es-Salaam City, Tanzania. Methodology: The study utilizes RLPs collected at the sub-ward level via the survey method and uses open data such as Nighttime Lights (NTL), and amenities coordinates points from OpenStreetMap. This paper explores the ability of two (2) ensemble machine learning methods (ie. Random Forest Regression (RF-R) and XGBoost Regression) for mapping RLP at grid-level. Results: Results found that RF-R was slightly superior to XGBoost Regression and was used to map RLP at fine-scale. The relative importance of explanatory variables in the RF-R model demonstrated that NTL was by far the most important determinant for the RLP spatial distribution in Dar-es-Salaam. NTL literature presents it as a proxy for socioeconomic variables such as Gross Domestic Product (GDP) and population, hence describing typical characteristics of informal land markets. Contrary to global-north urban studies with formal land markets whereby variables such as commercial and educational amenities are found to be very important in estimating RLPs. The paper presents a cost-effective methodological approach for mapping land prices at fine-scale in Dar-es-Salaam city and other cities with similar characteristics in the region, hence improving urban decision-making and policies.

Keywords

• residential land price
• land markets
• sub-saharan africa
• ensemble machine-learning
• urban
• mapping