Big Data Analytics @ Munich Re - Munich Re Life Forum · 10/20/2016 · Big Data Analytics @ Munich Re Munich Re Life Forum. ... HANA. User InterfaceUser Interface. SAS. A2P. ...

20 October 2016Wolfgang HaunerChief Data Officer, Munich Re

Bildquelle: Mark Moffett / Getty Images

Big Data Analytics @ Munich ReMunich Re Life Forum

Bildquelle: Mark Moffett / Getty Im

Agenda

20 October 2016 2Big Data Analytics @ Munich Re / Wolfgang Hauner

Data Analytics Framework1 Current Analytics Activities2

Method Example: AI3 Advanced Analytics: MR-Examples from the field4

© Munich Re

Loc-based services

Smart HomeTelematics

VirtualAssistantSystems

Haptic Technologies

Integrated Systems

Autonomous Systems and Devices

Automated Decision Taking

Cloud/Client ArchitectureNew Payment

Models

Big Data

Internet of Things

Cybersecurity

Digitalization

Computing Everywhere

Robotics/DronesWearable Devices

Risk-based Security

Context-aware Computing

Open Data

Collaborative Consumption

Predictive Analytics

Industrialization 4.0

Web 4.0Web-Scale IT

Software-defined Anything

Crowdsourcing

Mobile Health Services

3D Printing

Augmented and virtual worlds

Citizen Development

User Centered Design

Digital Identity

On-Demand-Everything

Big Data in Trend Radar


Big Data

Digitization

Internet of Things

© Munich Re

https://www.munichre.com/en/reinsurance/magazine/topics-online/2016/04/it-trends-2016/index.html

https://www.munichre.com/en/reinsurance/magazine/topics-online/2016/04/it-trends-2016/index.html

When does it become BIG Data?


43 zettabytes of data will probably be generated by 2020

300 times the volume in 2005

40,000,000,000,000,000,000,000

ByteKilobyteMegabyteGigabyteTerabytePetabyteExabyteZettabyte

Source: IBM

4 KB Commodore VC 203.5 inch floppy disk

Data contained in a library floor

4 TB in Memory Big Data Platform MR

Petabyte Storage Big Data Plattform

Google, Facebook, Microsoft…

All words ever spoken by humans

Yes or No

© Munich Re

Big Data Analytics

Methods Regression Models

Machine Learning Models

Text Mining

Technology Hardware

(Compute power)

Software (SAS, R, Spark, …)

Data Internal Data

External Data

Structured Data

Unstructured Data

People Data Scientists

Data Engineers

Business People

Big Data Analytics is a Combination of Methods, Technology, Data and People

520 October 2016Big Data Analytics @ Munich Re / Wolfgang Hauner© Munich Re

Building the Team, and the Environment

Programming

Story-telling

Statistics

Visualization

System Implemen-

tation

DB Administration

Maths

Modelling Data Storage

Business-/Domain

knowledge

20 October 2016Big Data Analytics @ Munich Re / Wolfgang Hauner 6

Business-Units IT

© Munich Re

Building the Infrastructure


SASHANA Hadoop Stack

HANA

User InterfaceUser Interface User Interface

SASHANA Hadoop Stack

HANA

User InterfaceUser Interface User Interface

A2P

Data Lake (HDFS)

Long term unstructured and structured data

BI Lab Production

© Munich Re

20 October 2016Big Data Analytics @ Munich Re / Wolfgang Hauner 8© Munich Re

Which topics drive our clients?

Up-/Cross-Selling

Data Sources

Textmining Churn Analysis

Supply Chain

Social Media Analysis

Fraud Detection

Big Data Technology

Predictive UW

Telematics

Sensor Data/IoT

Geospatial

Big Data Analytics @ Munich Re / Wolfgang Hauner 20 October 2016 9© Munich Re

Big Data use cases in insurance

Make the uninsurable insurable

Diabetics

Wind Energy

Consolidate the information and process

Automated underwriting

Risk management platform

Artificial Intelligence supported workflow

Early Loss Detection

Visual Loss Adjustment

Image: dpa Picture Alliance Image: Getty Images

Image: Getty Images

Image: used under license from shutterstock.com



Big Data Analytics @ Munich Re / Wolfgang Hauner 20 October 2016 10© Munich Re

Agenda




© Munich Re

Pilot Fact SheetInternet Research & Intelligence System (IRIS)


Multi-dimensional searches based on standardized search technology to accelerate web research (example Tianjin)

Extended analytics to gather further data insights, e.g., based on topic analysis and organizational grouping

Parallel processing and delta mechanism for multi-processed search requests

Results shown in different visualizations (word cloud, table, topic analysis, etc.) and exportable to Excel

Results Benefits

Outlook Additional analytics modules for better insights and

broader application Collaboration functionalities for more efficient

case analysis

© Munich Re

Text Mining and Web Crawling:Hong Kong Monetary authority announces FinTech ‘sandbox’

20 October 2016 13Big Data Analytics @ Munich Re / Wolfgang Hauner© Munich Re

OrganisationClients who are not active anymore are removed. The remaining data is split into INSURANCE and no INSURANCE

SeparationNow the data is randomly split into 5 even boxes. Each box contains both INSURANCE and no INSURANCE. However, the portion within each box varies.

TestingFor testing the first so called “set of training data” the first 4 boxes are aggregated again. Now they are used for sampling the first buying characteristics.

Random Forest (RF)Using machine learning methods, 300 decision trees will be generated simulating customer characteristics. Simulations show chains of combination for INSURANCE and no INSURANCE

ValidationThe just created random forest will now be used to back-test the remaining 5th

box: How accurate can we forecast who bought INSURANCE and who not?

?

Cross-Selling with Machine LearningAnalysis of pension, life and investment portfolios for product development and targeted sales

20 October 2016 14Big Data Analytics @ Munich Re / Wolfgang Hauner© Munich Re

Agenda




© Munich Re

Methods Neural Network Insurance specific Visual Intelligence


Insurance Companies, e.g., Munich Re, …

AI Community, e.g., Google, Facebook, …

Insurance specific Vision Intelligence

General ObjectVision Intelligence

Images left: used under license from shutterstock.comImage right: Getty Images

© Munich Re

System of interconnected nodes, exchanging information

Weights of connections can be adjusted by supervised/ unsupervised “learning”

Pros: Accuracy usually high, prediction fast

Cons: “Black box” – acquired knowledge not easily comprehensible, training effort high, appropriate data needed

Application areas, e.g., speech recognition, computer vision, medical diagnosis, automated trading, game-playing (AlphaGo)

MethodsNeural Network

17Big Data Analytics @ Munich Re / Wolfgang Hauner

Input Hidden Output



Image: Getty Images Image: Getty Images

No pothole identified


Pothole identified


No pothole identified

20 October 2016© Munich Re

MethodsPotential use-cases of Neural Network Infrastructure Insurance


Detect road damage

Categorize damage

Estimate claim

Trigger repair action

Image: used under license from shutterstock.com Image: used under license from shutterstock.com

© Munich Re

Agenda




© Munich Re

Remarks The set of explaining variables

differs based on the covers included (as expected)

Top 10 factors are mainly linked to accidental risk (occupation, activity, job position, free time activity). Explained by the high percentage of cases with accidental covers included

Predictive Underwriting with Machine LearningWhich factors explain the underwriting outcome, which are not significant?

20 October 2016 20

Only 20 from 58 fields are required to predict the underwriting result0 10 20 30 40 50 60 70 80 90 100

Occupation CodeQ: Sports

SubsidiaryBMI

Job activityCovers included

Job positionQ: Under treatment

Free time activityAge

Q: Systemdis./addict./scelet.GenderQ: Bike

Entry yearDiff. Age to partnerSum_insured_Life

Sum_insured_TRANSRelationship to benef. 1Relationship to benef. 2

Insurance cover code

Big Data Analytics @ Munich Re / Wolfgang Hauner© Munich Re

Remarks There are no questions which

are always answered with “YES” or “NO”

Some questions did not have any impact in the model (the data could not explain why)

Just because factors did not have any impact in the model didn’t mean the relating questions could be waived (impact on selection given, i.e., HIV question, rehabilitation for addiction) → careful consideration required

Predictive Underwriting with Machine LearningWhich application questions impact the underwriting outcome, which do not?

20 October 2016 21

Impact on probability for standard or loaded/rejected decisionCurrently doing dangerous sports?Currently under treatment or advised surgery?Internal disease, skeletal condition, addition?Cancer or neuro-psychol. condition in last 10y?Motorbike as competition?HIV/AIDS?Motorcross?Taken drugs in last 10y?Daily use of motorbike?Currently pregnant?Had treatment or medical exams in last 3y?Motorbike?Hospitalization in last 3y?Family history?Smoked in previous 12 months?Previous or advised rehabilitation for addition?Pregnancy complication?Stopped usual tasks in previous year?Plan to visit/reside abroad?

HighHigherLowLowerNo

Legend:


Compare different Machine Learning algorithms (Support Vector Machines, Random Forests, Boosted Trees, Regression Boosting, Lasso-regularized Regression) with classical GLMs

Applied to Mortality data

Additionally: Clear visualization of main and interaction effects

Machine Learning as alternative to classical modeling Get better performance + applicable to Big Data


Machine Learning helps in understanding and selecting the most relevant influential factors

© Munich Re

Observed and predicted average claimed amounts in 2012 itemized by age and gender (here only women)

Modern Machine Learning TechniquesAllows to detect more detailed patterns

23

GLM Random Forest

The traditional approach takes age and gender into account and therefore mostly performs quite good on average. Only the random forest detects the peak for women in their thirties (pregnancy treatments)

Traditional approach

20 October 2016Big Data Analytics @ Munich Re / Wolfgang Hauner© Munich Re

20 October 2016 24

Claim ID Description Topic 1 … Topic m Code

1 “Heart attack“ -0.35 … 0.64 08008

… … … … … …

n “Breast cancer“ 0.17 … -0.04 09999

Claim ID Description Topic 1 … Topic mPredicted

code

1 “Stroke“ -0.25 … -0.14 01999

… … … … … …

k “Ovarian cyst“ 0.81 … 0.63 04325

Claim IDPredicted

codeProba-

bility Check

1 01999 0.83

… … …

k 04325 0.27

Claims with codes – “training dataset” Machine learning model Rules

Claims without codes – “scoring dataset”Verify predictions on scoring dataset

MOCA (Medical and Occupational Coding Assistant)Tool that maps codes to medical descriptions


Digitization

Big Data

IoTBuilding the

Infrastructure

Building The Team

Building Business-

Cases

Consolidate Process & Information

A.I. supported Workflow

Make the Un-insurable Insurable

…

Big Data Trend is a fact,

bringing insurance industry challenges & opportunities …

→ engaging the trend properly

… to turn the challenges to business potential. Broad data sources

Advanced analytics

Visualization

ML/AI

Values for Insurance

From Trend to Business ValueAn on-going journey

20 October 2016Big Data Analytics @ Munich Re / Wolfgang Hauner 25© Munich Re

Thank you!

Contact: Wolfgang Hauner, Chief Data Officer, Munich [email protected]© 2016 Münchener Rückversicherungs-Gesellschaft © 2016 Munich Reinsurance Company

Image: Bayerische Zugspitzbahn Bergbahn AG / LechnerCenter for International Earth Science Information Network - CIESIN - Columbia University. 2016. Gridded Population of the World, Version 4 (GPWv4): Population Count. Palisades, NY: NASA Socioeconomic Data and Applications Center (SEDAC).

Big Data Analytics @ Munich Re - Munich Re Life Forum · 10/20/2016 · Big Data Analytics @ Munich Re Munich Re Life Forum. ... HANA. User InterfaceUser Interface. SAS. A2P. ...

Documents