Ubiquitous intelligence for smart cities: a public safety approach
Citizen-centered safety enhancement is an integral component of public safety and a top priority for decision makers in a smart city development. However, public safety agencies are constantly faced with the challenge of deterring crime. While most smart city initiatives have placed emphasis on the use of modern technology for fighting crime, this may not be sufficient to achieve a sustainable safe and smart city in a resource constrained environment, such as in Africa. In particular, crime series which is a set of crimes considered to have been committed by the same offender is currently less explored in developing nations and has great potential in helping to fight against crime and promoting safety in smart cities. This research focuses on detecting the situation of crime through data mining approaches that can be used to promote citizens' safety, and assist security agencies in knowledge-driven decision support, such as crime series identification. While much research has been conducted on crime hotspots, not enough has been done in the area of identifying crime series. This thesis presents a novel crime clustering model, CriClust, for crime series pattern (CSP) detection and mapping to derive useful knowledge from a crime dataset, drawing on sound scientific and mathematical principles, as well as assumptions from theories of environmental criminology. The analysis is augmented using a dual-threshold model, and pattern prevalence information is encoded in similarity graphs. Clusters are identified by finding highly-connected subgraphs using adaptive graph size and Monte-Carlo heuristics in the Karger-Stein mincut algorithm. We introduce two new interest measures: (i) Proportion Difference Evaluation (PDE), which reveals the propagation effect of a series and dominant series; and (ii) Pattern Space Enumeration (PSE), which reveals underlying strong correlations and defining features for a series. Our findings on experimental quasi-real data set, generated based on expert knowledge recommendation, reveal that identifying CSP and statistically interpretable patterns could contribute significantly to strengthening public safety service delivery in a smart city development. Evaluation was conducted to investigate: (i) the reliability of the model in identifying all inherent series in a crime dataset; (ii) the scalability of the model with varying crime records volume; and (iii) unique features of the model compared to competing baseline algorithms and related research. It was found that Monte Carlo technique and adaptive graph size mechanism for crime similarity clustering yield substantial improvement. The study also found that proportion estimation (PDE) and PSE of series clusters can provide valuable insight into crime deterrence strategies. Furthermore, visual enhancement of clusters using graphical approaches to organising information and presenting a unified viable view promotes a prompt identification of important areas demanding attention. Our model particularly attempts to preserve desirable and robust statistical properties. This research presents considerable empirical evidence that the proposed crime cluster (CriClust) model is promising and can assist in deriving useful crime pattern knowledge, contributing knowledge services for public safety authorities and intelligence gathering organisations in developing nations, thereby promoting a sustainable "safe and smart" city.