Background & context
For the Smart Policing Seoul Center initiative, the goal was to predict which areas carry elevated crime risk so policing resources can be allocated proactively.
That meant turning raw crime and location data into spatial features a model could learn from.
Key responsibilities & role
- Geospatial featuresUsed QGIS and GeoPandas to engineer spatial features from crime and location data.
- DataManaged the spatial dataset in PostgreSQL.
- ModelingTrained a LightGBM model to predict high-risk areas.
- IterationTuned features and the model to raise predictive accuracy.
Tools & knowledge
PythonQGISGeoPandasPostgreSQLLightGBM
Results & achievements
→ Achieved 80% prediction accuracy on high-risk-area classification.
→ Delivered for the Smart Policing Seoul Center initiative.