Beyond linear models: machine learning insights into the determinants of slum prevalence

Résumé

Introduction Rapid urbanization in developing countries has intensified the growth of informal settlements, raising critical concerns for housing affordability and sustainable urban planning. Despite extensive research, the drivers of urban slum incidence remain complex and potentially nonlinear, requiring more advanced analytical approaches. Methods This study examines the nonlinear determinants of urban slum incidence using a cross-country dataset of 436 country-year observations derived from the World Bank’s World Development Indicators. A Random Forest regression model is employed to capture complex relationships between slum prevalence and a broad set of economic, demographic, infrastructure, education, and health variables. A multiple linear regression model is also estimated as a benchmark for comparison. Results The findings reveal that infrastructural factors, particularly access to sanitation, drinking water, and electricity, are the most significant predictors of slum incidence. Demographic pressures, including population growth and density, further exacerbate slum conditions by increasing housing demand and straining urban systems. In contrast, economic indicators such as GDP per capita and economic growth exhibit weaker and less consistent effects. The Random Forest model outperforms the linear regression model (R 2 = 0.822 vs. 0.661), indicating superior predictive performance. Discussion The results shed light on the critical role of basic service provision and housing quality in reducing slum prevalence, suggesting that economic growth alone is insufficient without corresponding infrastructural development. Furthermore, the superior performance of the Random Forest model highlights the importance of nonlinear analytical approaches in capturing the complexity of urban dynamics and informing more targeted and effective policy interventions.

Citer ce document

Accès au document

Texte intégral en lecture en ligne, réservé aux abonnés SPHAERO et aux membres de l'institution. Se connecter

Auteur(s)

Statistiques

Consultations : 1

Téléchargements : 0