A systematic review on spatial crime forecasting.

Area of Science:

• Criminology
• Data Science
• Geographic Information Systems (GIS)

Background:

• Predictive policing and spatiotemporal crime analytics are gaining traction in scientific and law enforcement communities.
• These tools are increasingly implemented for effective policing strategies.

Purpose of the Study:

• To provide an overview and evaluation of the state-of-the-art in spatial crime forecasting.
• Focus on study design and technical aspects of existing research.

Main Methods:

• Systematic literature review following PRISMA guidelines.
• Analysis of 32 papers published between 2000 and 2018.
• Papers selected based on publication type, relevance, and study characteristics.

Main Results:

• Hotspots (binary classification) are the predominant forecasting method.
• Traditional machine learning, kernel density estimation, point process, and deep learning approaches were utilized.
• Prediction Accuracy, Prediction Accuracy Index, and F1-Score are key evaluation metrics.
• Train-test split was the most common validation approach.

Conclusions:

• Spatial crime forecasting shows significant growth due to interdisciplinary collaboration.
• Studies aim to address societal needs in crime understanding and law enforcement's real-time prediction interests.
• Inconsistent reporting, lack of experimental detail, and varied terminology present limitations.
• Standardization of approaches and reporting key data items are suggested for improved comparison and reproducibility.