Using AI to Prevent Residential Burglary in Berlin
summary
Krimpro is a predictive system deployed in Berlin since 2017 that forecasts residential burglary risk per 400×400m quadrant for the next 72 hours, trained on ~50,000 crimes (≈7.3M quadrant-days) and built in ~100 development days; models run on-premise using decision trees / a small random forest on a 2-core 16GB server, use no personal data, cluster predictions into patrol-sized areas, and deliver about a ~60% hit rate in operational use, relying heavily on police expertise and public data.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:10About Aurelius
- 3:55Near Repeat Theory
- 5:44Solution
- 9:38Heatmaps
- 12:17Development
- 13:38Results
- 15:37Success Factors
- 18:25Questions
related talks
16mDeep Tech From Berlin & Novi Sad
22mUrban Artificial Intelligence in Amsterdam
44mPlatform Intrusion Detection with Deep Learning
26mSridhar Telidevara: Professor, Great Lakes Institute of Management at Cypher 2017 - Part 1/2
08mAI-Powered CRM in Policing: Innovation and Surveillance
25mReal-Time Fraud Detection at Scale
01h 06mAmazon Machine Learning to Predict Responses
37mAI for Good- Then and Now