African countries follow distinct slum prevalence trajectories, clustering analysis reveals

Progress toward the United Nations housing target SDG 11.1 in Africa cannot be read from a single year’s ranking, according to a new study that treats each country’s slum prevalence as a 23-year trajectory rather than a snapshot. The research, published in Frontiers in Built Environment, applied hierarchical clustering to UN-Habitat data covering all 54 universally recognised sovereign African states from 2000 to 2022, and found that the continent does not follow one common temporal pattern. Instead, countries fall into two broad trajectory groups distinguished by their relative position within the annual African distribution, yet the countries inside each group differ markedly in how their situations changed over time.

The study used the UN-Habitat Urban Indicators Database, specifically the dataset titled “Housing, slums and informal settlements”, Version May 2024, downloaded on 12 April 2026. The indicator measures the proportion of the urban population living in slums or informal settlements, which the authors treat as one housing-related component of SDG indicator 11.1.1. They are careful to note that the full indicator also covers inadequate housing, and that slum prevalence should not be read as a complete measure of housing deprivation or of SDG 11 as a whole. The published UN-Habitat series was retained as supplied, because it draws on national and international statistical information and harmonised estimation procedures rather than a single survey instrument applied uniformly across all countries and years.

A crucial methodological choice shapes what the clusters mean. Before clustering, the researchers standardised the prevalence values within each year across the 54 countries, so that each country-year observation expresses a position relative to the contemporaneous African mean. A positive value indicates prevalence above that year’s continental average, a negative value below it. These transformed values are not percentages and do not represent percentage-point changes. As a result, the analysis identifies similarities in relative prevalence trajectories rather than absolute prevalence levels. A country can remain relatively high within the continental distribution while its absolute prevalence changes considerably, so the clusters are better interpreted as patterns of persistent relative positioning than as fixed levels of housing deprivation.

Each country was represented as a 23-dimensional vector of standardised annual values, giving a balanced dataset of 1,242 country-year observations. The primary specification applied Ward’s minimum-variance linkage with Euclidean distance across the complete ordered trajectory, preserving the chronological sequence of observations rather than reducing them to starting values, endpoints or average trends. The number of clusters was assessed across solutions from two to ten using four internal diagnostics: the silhouette coefficient, the Calinski-Harabasz index, the Davies-Bouldin index and the within-cluster sum of squares. The two-cluster solution scored highest on the silhouette measure at 0.4823, compared with 0.4135 for three clusters and 0.3796 for four, and recorded a Davies-Bouldin value of 0.7734. A multi-criterion ranking that combined these measures with bootstrap stability selected two clusters as the preferred solution.

Bootstrap resampling tested how consistently the partitions reappear when countries are resampled. Five hundred replications were run for each candidate solution, and agreement with the original clustering was quantified with the Adjusted Rand Index. The two-cluster solution produced a mean bootstrap ARI of 0.6464, with a 95% percentile interval of 0.1275 to 1.0000. Notably, the four-cluster solution was more stable, with a mean ARI of 0.7405, but showed weaker separation on the silhouette and Davies-Bouldin measures. The authors interpret this contrast as informative rather than contradictory: two clusters capture the broadest distinction in relative trajectory position, while a finer partition reveals additional internal structure within those broad groups.

The preferred solution assigned 34 countries, about 63% of the sample, to a cluster with generally higher relative prevalence throughout the period, and 20 countries, about 37%, to a cluster with generally lower relative prevalence. The higher-prevalence group spans West, Central, East and parts of Southern Africa and includes countries as different as Nigeria, Kenya, Mauritius, Cabo Verde and Seychelles. The lower-prevalence group includes North African countries such as Egypt, Libya, Morocco and Tunisia alongside South Africa, Botswana, Namibia, Ghana, Senegal, Angola, Rwanda and Zimbabwe. Neither cluster corresponds closely to a conventional geographical division, suggesting that geographical proximity alone does not account for the temporal similarities in the data.

The mean trajectories of the two groups remained separated across the entire period. The higher-prevalence cluster averaged standardised values of 0.65 in 2000 and 0.62 in 2022, while the lower-prevalence cluster averaged −0.89 and −0.75 in those years. The gap between the cluster means narrowed only modestly, from 1.54 standardised units in 2000 to 1.37 in 2022. The higher-prevalence group had a descriptive slope of −0.003 and the lower-prevalence group a slope of 0.008, with volatility values of 0.009 and 0.011 respectively. Djibouti and Ghana served as the cluster medoids, the actual country trajectories closest to the average pattern of their groups.

Within-group heterogeneity is where the analysis becomes most revealing. Angola recorded the strongest positive slope among the reported countries at 0.173, with a total change of 2.940 standardised units, while Burundi showed a strong negative slope of −0.070 and a total change of −1.237. Burkina Faso displayed the highest volatility at 0.602, whereas Botswana moved comparatively little, with a slope of 0.001 and a total change of −0.065. The Democratic Republic of the Congo and Equatorial Guinea also recorded notable positive slopes of 0.064 and 0.066. Principal component analysis supported the visual separation of the clusters, with the first two components explaining 88.5% and 8.3% of the variance in the trajectories, a combined 96.8%.

Sensitivity analyses complicate the picture in an important way. When the trajectories were re-represented as first differences, emphasising year-to-year movements, or clustered with correlation distance, emphasising trajectory shape, the preferred number of clusters remained two, but agreement with the primary country assignments collapsed, with ARI values of −0.037 and −0.006 respectively. The number of broad groups is therefore robust, while individual country membership depends heavily on which aspect of temporal behaviour is given the greatest weight. The authors stress that the clustering is descriptive, not causal: it identifies patterns of similarity among observed trajectories and does not estimate the effects of economic, demographic, institutional or policy variables.

The study’s implications for SDG 11.1 monitoring follow directly from these findings. Annual rankings capture relative position but can conceal persistence, reversal or volatility, so trajectory-based assessment can complement conventional reporting by distinguishing sustained improvement from recent recovery or deterioration. The authors also caution against reading the lower-prevalence cluster as evidence of policy success, since the indicator covers slums and informal settlements rather than all dimensions of housing adequacy. They note that several countries changed their relative positions in the later years of the series, coinciding with a period of global disruption, but the clustering results do not establish that COVID-19 or any other external event produced these movements. The analysis identifies temporal patterns; explaining why countries follow different housing pathways requires separate longitudinal investigation of governance, income, urbanisation, infrastructure and housing policy.

Subject of Research: Trajectory clustering of slum and informal-settlement prevalence across African countries from 2000 to 2022

Article Title: Uneven progress toward SDG 11.1 in Africa: a hierarchical clustering analysis of slum prevalence trajectories

Article References: Ametepey, S. O., Aigbavboa, C., & Ofori, K. (2026). Uneven progress toward SDG 11.1 in Africa: a hierarchical clustering analysis of slum prevalence trajectories. Frontiers in Built Environment, 12, Article 1917266. https://doi.org/10.3389/fbuil.2026.1917266

Image Credits: AI Generated

DOI: 10.3389/fbuil.2026.1917266

Keywords: slum prevalence, SDG 11.1, hierarchical clustering, UN-Habitat, informal settlements, Africa, urban housing, trajectory analysis, longitudinal data, urbanisation, cluster stability, housing policy