SAFARI Situational Awareness Framework for Risk Ranking Alberto - - PowerPoint PPT Presentation

MIT Geospatial Data Center

SAFARI: Situational Awareness Framework for Risk Ranking

1

SAFARI

Situational Awareness Framework for Risk Ranking

Alberto Garcia-Robledo, Abel Sanchez, Rongsha Li, Juan-Carlos Murillo-Torres, John Williams and Sascha Boheme Massachusetts Institute of Technology MIT Geospatial Data Center

z

Challenges of Fraud Detection

• Few samples of fraud: i.e. unlabeled data (no supervised learning)
• Anomaly detection (unsupervised learning): high false positive rate
• Results may make no sense: what do the outliers actually mean?
• Reducing dimensions causes that data to lose its semantic meaning
• Finding a needle in a haystack

How we can best exploit the unlabeled payment transaction douments considering the different types of information contained in them?

MIT Geospatial Data Center

SAFARI: Situational Awareness Framework for Risk Ranking

2

Situational Awareness for Fraud Detection

SAFARI introduces the concept of Situational Awareness to enable the detection of fraud on large volumes of payments where ground truth is not available, by integrating different perspectives of financial data.

MIT Geospatial Data Center

SAFARI: Situational Awareness Framework for Risk Ranking

3

Level 3:

Projection

Situational Awareness

SA Risk Managment and Prediction Data Collection and Anomaly Detection Visual Analytics and Data Integration Predictive Analytics

Level 2:

Comprehension

Level 1:

Perception

RF raising at different perspectives

• f data

RFNet modelling and Web dashboard visualizations

SAFARI: 1 and 2-Level SA Approach

MIT Geospatial Data Center

SAFARI: Situational Awareness Framework for Risk Ranking

4

SAFARI: Workflow

PRC Payment Docs. GXM Payment Docs. GAX Payment Docs.

...

MIT Geospatial Data Center

SAFARI: Situational Awareness Framework for Risk Ranking

5

SAFARI: Data Ingestion and Enrichment

Enricher

MIT Geospatial Data Center

SAFARI: Situational Awareness Framework for Risk Ranking

6

N-way

SAFARI: Anomaly Detection

N-way

N-way Exact Matching

N-way N-way

String Fuzzy Matching

N-way N-way

String Phonetic Matching

N-way N-way

Address Geolocation Matching

SAFARI: Situational Awareness Framework for Risk Ranking

7 Accenture | MIT Alliance in Business Analytics

N-way N-way

Rule Expression Matching

MIT Geospatial Data Center

SAFARI: Situational Awareness Framework for Risk Ranking

7

SAFARI: RFs and RFNet Integration

RFNet

MIT Geospatial Data Center

SAFARI: Situational Awareness Framework for Risk Ranking

8

P R Q O N F G H J K L M A B C D E I

RFNet 2 RFNet 1 RFNet 3 RFNet 5 RFNet 4

SAFARI: RFNet Scenarios

MIT Geospatial Data Center

SAFARI: Situational Awareness Framework for Risk Ranking

9

RFNet Pair-Wise Matching and BBN Ranking

A B

FuzzyA PhonA GeoA

A B

r = 0.95

VendorMatch r NameMatch e AddressMatch f NamePhonMatch b AddressGeoMatch c NameFuzzyMatch

a

AddressNwayMatch d

P(r = T) = 0.95 P(a = T) = 1 P(b = T) = 1 P(c = T) = 1 P(d = T) = 0 P(e = T) = 1 P(f = T) = 0.8 MIT Geospatial Data Center

SAFARI: Situational Awareness Framework for Risk Ranking

10

SAFARI: Risk Score Propagation

C G A E D F 0.95 0.84 0.91 0.65 0.65 0.23 0.23 B

C G

A

E D F 0.95 0.84 0.91 0.65 0.65 0.23 0.23

B

0.95 0.95 0.91 0.91 0.84 0.23 0.65

RFNet score: 0.95

MIT Geospatial Data Center

SAFARI: Situational Awareness Framework for Risk Ranking

11

P R Q O N F G H J K L M

A

B

C

D E I

Risk: Low

Risk: High

Risk: Low Risk: Low Risk: Low

SAFARI: Finding the Needle ...

MIT Geospatial Data Center

SAFARI: Situational Awareness Framework for Risk Ranking

12

P R Q O N F G H J K L M

A

B

C

D E I

Risk: Low

Risk: High

Risk: Low Risk: Low Risk: Low

SAFARI: Finding the Needle ...

MIT Geospatial Data Center

SAFARI: Situational Awareness Framework for Risk Ranking

13

SAFARI: Web-Based Visual Analytics

MIT Geospatial Data Center

SAFARI: Situational Awareness Framework for Risk Ranking

14

Conclusions

• Analysis integration. Combine different analysis

techniques for processing large amounts of payment documents.

• Big data analysis. Help SMEs to make sense of a large

amount of RFs spread across data.

• Focus. Help SMEs to focus on the most suspicious payments

by exploiting modern high-performance multi-core computers and visualization techniques.

False positive minimization

Novelty:

Integration Ranking Visualization

MIT Geospatial Data Center

SAFARI: Situational Awareness Framework for Risk Ranking

15