SLIDE 1
(Machine) Learning To Detect Fraudsters
Hany Elemary Sarah LeBlanc
CREDIT CARD FRAUD
2
(Machine) Learning To Detect Fraudsters Hany Elemary Sarah LeBlanc - - PowerPoint PPT Presentation
(Machine) Learning To Detect Fraudsters Hany Elemary Sarah LeBlanc - - PowerPoint PPT Presentation
(Machine) Learning To Detect Fraudsters Hany Elemary Sarah LeBlanc CREDIT CARD FRAUD TRANSACTION APPLICATION CARD NOT FOUND 2 FRAUD DETECTION MODEL PLOT Not Fraud Fraud Application Count False Negatives False Positives 0 0.2 0.4
SLIDE 2
SLIDE 3
FRAUD DETECTION MODEL PLOT
3
1 0.2 0.4 0.6 0.8
Model Score Application Count
Not Fraud Fraud False Positives False Negatives
SLIDE 4
MINIMIZE LOSSES
4
Lost Profitability = (Fraud Cost * FN) + (Opportunity Cost * FP)
FP (Mistaken fraud) FN (Fraud missed) Legend:
SLIDE 5
Fraud Detection
Application Service
CURRENT STATE
5
Vendor
Customer
Rules Model Strategies
SLIDE 6
Fraud Detection Customer
Vendor
Application Service
PROPOSED STATE
6
Fraud Detection
Strategies Rules
CHALLENGER MODELS CHAMPION
SLIDE 7
MODEL TRAINING
7
Historical Data Training Model Classification Fraud Not Fraud
Supervised Learning
SLIDE 8
8
DATA PATTERNS
Filter Transform Impute Features
SLIDE 9
9
DATA FILTERING
Low Cardinality
SLIDE 10
10
DATA FILTERING
High Cardinality
SLIDE 11
11
DATA FILTERING
Medium Cardinality
SLIDE 12
12
DATA FILTERING
Predictive Model Training
Medium Cardinality
SLIDE 13 Email
DATA TRANSFORMATION
13
Fraud Status
jack.smith@gmail.com annie.may@fraudster.com freddy.jr@gmail.com nicole.jack@fraudster.com jon.johnston@gmail.com claudia.penns@us.gov walter.carson@gmail.com ben.benjamin@fraudster.com
SLIDE 14
DATA TRANSFORMATION
14
Domain name Fraud Status
gmail.com fraudster.com gmail.com fraudster.com gmail.com us.gov gmail.com fraudster.com
SLIDE 15
15
DATA IMPUTATION
Column 1
Column 2 Column 3 Column 4
Handling Missing Data
SLIDE 16
16
DATA IMPUTATION
Column 1
Column 2 Column 3 Column 4
Handling Missing Data
SLIDE 17
17
FEATURE SELECTION
IP to Zip Proximity
SLIDE 18
ARCHITECTURE
18
DATA SCIENTIST WORKFLOW DEVELOPER WORKFLOW Transformed Data Trained Model Raw Data Trained Model Score Applications
SLIDE 19
Transformed Data Raw Data
19
DATA SCIENTIST WORKFLOW
Clean Impute Transform
Binary Repository Historical Data Store
Trained Model
SLIDE 20
Trained Model Score Applications
Message Queue
20
DEVELOPER WORKFLOW
Model 1 Model 2 Model 3
Shadow Mode
Decisioning & Analytics Platform Application Service
Vendor
Model Predictions Store
Rules Model Strategies
Binary Repository
Model 1
SLIDE 21
21
Decisioning & Analytics Platform Application Service
Model Predictions Store
Message Queue Model 1 Model 2 Model 3
Shadow Mode
Champion Model
Binary Repository
DEVELOPER WORKFLOW
Vendor
Rules Strategies
Trained Model Score Applications
SLIDE 22
22
DEVELOPER WORKFLOW
DATA SCIENTIST WORKFLOW
ARCHITECTURE
Rul Str
SLIDE 23
23
VALUE STREAM
Data Ingestion Model Training Governance Publish Model Publish Service Shadow Mode Governance Evaluation Champion Model
25 50 75 100 SLIDE 24
24
THANK YOU
Sarah LeBlanc Hany Elemary
sleblanc@thoughtworks.com helemary@thoughtworks.com @sarah_g_leblanc @hanyelemary