Mapping Informality and Violence: Machine Learning Insights into Crime Patterns Across South African Police Jurisdictions

This study examines the relationship between neighborhood informality and crime rates in South Africa, where rapid urbanization has led to criminal organizations stepping in to provide services within some of these settlements. Applying K-means clustering and Local Moran's I analysis to 1,164 police jurisdictions, we observe that the cluster with the highest level of informality also exhibited the lowest crime rates. Furthermore, in clusters representing moderately dense urban areas, linear regression reveals a statistically significant negative correlation between levels of informality and crime rates. Machine learning models, including KNN and random forest, show that crime can be predicted with low MSE, with k-complexity emerging as a key feature. Finally, a longitudinal case study on crime and growth data offers further insight into potential causal relationships, though the results are not statistically significant. These findings have implications for urban governance and public safety policies in developing countries.