Browsing by Author "Macingwane Apiwe"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Investigating Frequent Pattern-based Models for Im- proving Community Policing in South Africa
    (University of te Western Cape, 2025) Macingwane Apiwe
    South Africa (SA) faces significant challenges due to a high crime rate, posing threats to citizens’ safety and economic growth. Adopting a comprehensive approach that integrates traditional crime prevention methods with advanced technologies is essential to address these challenges effectively. One of the relevant advanced techniques is pattern mining in Data Mining (DM), which utilizes association rules to identify relationships within dataset variables. This method is pivotal in the crime sector as it mines frequent patterns associated with prevailing criminal activities, aiding law enforcement in making strategic decisions to combat crime. Despite limited research on pattern-based models in SA, Frequent Pattern Growth (FP-Growth) and Hyper Structure Mining (Hmine) models have proven to be effective and efficient across various fields. Therefore, this research delves into exploring two prominent frequent pattern-based mining algorithms, FP-Growth and Hmine, and then proposes a novel Hybrid Pattern- Growth algorithm (HP-Growth), which combines the strengths of FP-Growth and Hmine. An experiment was conducted with these models using Python to generate frequent patterns of crime with the South African crime statistics (Stats SA crime) dataset between 2005 and 2016 across all nine provinces. The experiment employed the Mean and Floor functions to compute the Minimum Support Value (MSV) required for pattern generation. Each model was evaluated using standard metrics such as memory usage, runtime, scalability, and the reliability of their frequent patterns. The study found that among the three models, HP-Growth performed more efficiently with sparse datasets, while Hmine performed best with dense datasets. FP-Growth exhibits higher time complexity and memory usage compared to Hmine and HP-Growth. Thus, Hmine emerges as the preferred algorithm for pattern generation due to its speed and memory efficiency with dense datasets, making it ideal for resource-constrained environments. The study establishes association rule thresholds and emphasizes the importance of selecting the most appropriate pattern-based model with low time and memory complexity for large-scale datasets and real-time processing for crime knowledge support. Furthermore, this research developed a crime knowledge support system, called CrimeTracker, which allows users to report and view crime information, and receive crime alerts. The system integrates the most suitable pattern-based model using a Representational State Transfer (REST) Application Programming Interface (API). The development of CrimeTracker utilized Hypertext Markup Language (HTML), Cascading Style Sheets (CSS), Python, and JavaScript. The research findings on the developed CrimeTracker are intended to have far-reaching implications for law enforcement agencies and crime analysts in SA. These insights could be crucial in guiding efficient resource allocation, refining crime prevention strategies, and bolstering community empowerment for improved safety and social cohesion. This further aligns with Sustainable Development Goal (SDG) 16 on peace, justice, and strong institutions.