This dataset contains the structured data extraction matrix used in the systematic literature review titled “Spatio-Temporal Prediction of Urban Theft and Robbery Using Artificial Intelligence: A PRISMA-Based Systematic Review.” The dataset compiles information from 51 peer-reviewed scientific publications published between 2020 and 2025 that investigate artificial intelligence approaches for the spatio-temporal prediction of theft and robbery crimes in urban environments. The dataset was constructed following the PRISMA 2020 guidelines for systematic literature reviews. Each record represents a scientific study and includes detailed variables describing: • Bibliographic information (authors, year, journal, publisher)• Indexation status (Scopus, Web of Science, or other sources)• Geographic region of the study• Crime type and prediction target• Spatial and temporal modeling characteristics• Dataset sources and sizes• Artificial intelligence approaches (ML, DL, or hybrid methods)• Specific predictive models used• Feature types employed in the models• Evaluation metrics and best reported performance• Validation strategies• Key findings and main limitations reported by the authors The dataset is intended to support reproducibility, transparency, and future research on AI-based crime prediction and urban security analytics. This dataset accompanies the systematic literature review manuscript currently under preparation and provides the full extraction matrix used in the qualitative synthesis.