Analyzing Crime Hotspots in Buenos Aires Using Spatial Statistics and Machine Learning

This project targets the Palermo district of Buenos Aires, combining spatial statistics and machine learning algorithms to build a crime analysis tool. It provides urban planners and security departments with visual crime hotspot prediction and risk area identification, supporting data-driven urban safety decision-making.

In modern urban management, crime prevention and safety planning increasingly rely on data science. Traditional crime statistics can only present historical case locations and cannot predict future risks. The combination of spatial statistics and machine learning provides a new perspective for security departments—predicting high-risk areas by analyzing spatial patterns in historical data. This open-source project focuses on the Palermo neighborhood, converting geospatial data into intuitive visual results to help decision-makers scientifically allocate police forces and plan safety facilities.

The core value of the project lies in encapsulating complex technical processes into a user-friendly tool, allowing users without programming backgrounds to obtain professional analysis results through interface operations. The system supports three mainstream machine learning models: XGBoost (excellent gradient boosting capability), Random Forest (ensemble trees enhance stability), and Logistic Regression (simple and interpretable benchmark). Users can select and compare their effects. Visualization uses color-coded maps: red for high probability, green for safe areas, and blue for marking recent cases, enabling non-technical users to quickly understand.

The tool relies on the R language data science ecosystem: caret and tidymodels are used for model training and tuning; ggplot, tmap, and leaflet for map visualization; sf and sp for geospatial data conversion and computation. The architecture hides technical complexity behind a simple interface while providing complete source code and documentation for technical personnel to learn spatial statistics and geographic visualization.

The tool targets a wide range of users: urban planning departments to optimize public lighting and monitoring layouts; police departments to adjust patrol routes and police deployment; community organizations to carry out targeted safety publicity; ordinary residents to understand the safety status of their surroundings. Data-driven decision-making breaks the reliance on experience and improves the scientificity and fairness of urban safety governance.

The system supports exporting reports in multiple formats: saving map screenshots/PDFs for presentations, exporting CSV raw data for further analysis. Reports provide statistical information segmented by neighborhood, time period, and crime type, helping to understand crime patterns from multiple dimensions. Flexible output is suitable for real-time monitoring, regular safety assessments, and long-term trend analysis. Data accumulation forms time series to support in-depth analysis such as seasonal changes and policy effect evaluation.

The current project focuses on the Palermo area with limited data coverage; applications in other cities require users to prepare their own data and adjust boundaries. Crime prediction is sensitive, and results should be used as a reference rather than the only basis to avoid unfairness caused by algorithmic bias. Outlook: Open-source technology has great potential in social governance. With more cities opening up data and developers participating, the tool may evolve into a universal urban safety analysis platform, providing globally reusable solutions.