Presentation is loading. Please wait.

Presentation is loading. Please wait.

Going Beyond Averages, Using Spatial Data to Analyze Insurance Risk Scott Tracy, QPC Jennifer Lemus, ISO Innovative Analytics David Lapp, Farallon Geographics.

Similar presentations


Presentation on theme: "Going Beyond Averages, Using Spatial Data to Analyze Insurance Risk Scott Tracy, QPC Jennifer Lemus, ISO Innovative Analytics David Lapp, Farallon Geographics."— Presentation transcript:

1 Going Beyond Averages, Using Spatial Data to Analyze Insurance Risk Scott Tracy, QPC Jennifer Lemus, ISO Innovative Analytics David Lapp, Farallon Geographics Location Intelligence, San Francisco, 4/2006

2 What determines how much you pay for Auto Insurance?  Personal History (Accidents/Violations)  Type of Vehicles  Driving Conditions  Historically, these have been evaluated by averaging the loss experience in a given geographic area (typically a collection of zip codes)

3 Issues with Current Insurance Practices  Zip codes were constructed for the convenience of the Post Office  Zip codes are not homogenous when it comes to insurance risk, demographics

4 San Francisco Rating Territories

5 ZIP Code 94109: A Tour The Tenderloin: "...the haunt of the low and vile of every kind. ….Licentiousness, debauchery, pollution, loathsome disease, insanity from dissipation, misery, poverty, blasphemy and death are there. And Hell, yawning to receive the putrid mass, is there also. “ Robert Louis Stevenson declared "Nob Hill, the Hill of Palaces, must certainly be counted the best part of San Francisco." Japantown Fishermen’s Wharf

6 Factors that affect Driving Conditions  Businesses  Street Types  Use of Mass Transit  Weather  Commute Patterns  Population Density

7 Geoprocessing Requirements Past: ~300 parameters per Block Group... ~60M values Block Groups (~200K) Number of features per type (SIC, CFCC etc) per distance Distance to nearest per type Elevation statistics (avg, var, min, max) Nearest traffic data Business locations (~3M) Landmark locations (~1M) Traffic data locations (~1M) Elevation data (~250M) Present: ~500 parameters per Block Group...~100M values Future: Process ~20M Policy Locations... Billions of values! QPC predictive analysis Traditional non location- based analytic data

8 Technology evaluation Technical Requirements  Scalability  Performance  Flexibility Selected Oracle 10g Spatial for geoprocessing Server-side processing Extremely scalable to support expected growth Interoperable with GIS when functionality becomes necessary Technology Options  In-house software development  GIS  Spatial RDBMS

9 Geoprocessing overview Oracle 10g Spatial GIS platforms Web mapping appsSQL/Java API Nation-wide Source Locations (Census Block Groups, Policy Locations etc) Nation-wide Target Locations (Businesses, Landmarks, Traffic, Elevation etc) Geoprocessing results Standard Oracle development best practices applied to geoprocessing: Indexing (incl Spatial), Partitioning (incl Spatial), and many other optimization techniques Rapid construction of spatial data warehouse ~400 location-based values / sec ~30M location-based values / day (on less than state-of-the art h/w) QPC analysis platforms

10 Test bed geoprocessing Asynchronous complete geoprocessing Results review and extraction for SAS etc Geoprocessing User Interface

11 Effect of Different Traffic Generators Top TierBottom Tier Restaurant30%Racetrack or amusement park 11% Grocery Store26%Hotel, motel resort, or spa5% Elementary or Secondary School 26%National park or forest4% Bank25%Local or community park3% Car Dealer23%Airport2% Gas Station22%Doctor’s office or clinic1% Liquor Store18%Religious institution-10% Increase in physical damage claims by living within one mile of:

12 Zip Code 94106: Model Differentiation

13 Improve over current practices Current Practices IIA Model % Improvement Bodily Injury16.47%20.13%22% Physical Damage 8.28%11.83%42% Collision9.34%11.28%20% Comprehensive17.66%20.07%13% Gini Indexes (representing gains over random model)

14 Thanks for attending! stracy@qualityplanning.com jlemus@qualityplanning.com dlapp@fargeo.com


Download ppt "Going Beyond Averages, Using Spatial Data to Analyze Insurance Risk Scott Tracy, QPC Jennifer Lemus, ISO Innovative Analytics David Lapp, Farallon Geographics."

Similar presentations


Ads by Google