NO1 Certified Black Magic Specialist Expert Amil baba in Lahore Islamabad Raw...
Detect fraudulent credit card cash outs with graph-based methods
1. Perfecting IT service and favoring clients 'success
Fraudulent credit card cash-out detection
On Graphs
2. Fraudulent Credit Card Cash Out
The prevalence of Fraudulent credit card cashing out has led to a rise in the charge-off and delinquency rate
of credit cards, meanwhile the joint debt risk is transmitted to banks.
Bank 2014 2015 2016 2017 2018 2019
Construction
Bank
0.85% 1.08% 1.28% 1.17% 1.09% 1.21%
Bank of
Communication
1.68% 1.82% 2.14% 1.98% 1.84% 2.49%
China
Merchants Bank
0.94% 1.37% 1.21% 1.26% 1.14% 1.30%
China CITIC
Bank
- - 1.41% 1.30% 0.98% 1.74%
Ping An Bank 2.77% 2.5% 2.15% 1.20% 1.19% 1.37%
The charge-off and delinquency rate of major banks
in the past five years
3. Credit card fraud
Credit card fraud detection
Graph Searching
Rating Scale
Multipartite graph
Application Control
Customer Rating
Anti – Cash Out
Anti - Fraud
Transaction Early
Warning
Technology Application
Feature Engineering
Early stage
of Loan
Middle and Late
Stage of Loan
Machine Learning
Graph Embedding
Credit card fraud Scene
Skimming
Account takeover
Fake card
Application Fraud
Fraudulent Cash out
5. Data Problem of Bank
Massive Bank
Statements
Perfect Bills Data Island
Scarce Bad
Samples
6. Graph Schema
Building heterogeneous information network based on a large size of transaction records.
Vertic
es
• Clients
• Cards
• Business fields
Edges
• Relationship
7. Credit Card
Debit CardConsume
Payment
Store
The aim of detection is to find out the loop of
credit card funds inflows and outflows from
the graph.
Rules of fraudulent credit card cash-out detection
Revolving Credit: High frequency of transaction;
Consumer Credit: Installment credit, large transaction.
8. 1. Improve detection effect ;
--Fund Recycle Detection
2. Improve iteration efficiency;
--28 Times
3. Improve coverage efficiency
--77%.
Result Based on Rules Display
Business Interview,
Rules Combing
Sample Analysis,
Result comparison
Implement Rules
78 rules in total; 8 main cash-out methods
9. Graph-based transaction detection
We apply densest subgraph detection for cash-out transaction.
2.Credit card and debit card transfer
data
1.Detect fraudulent cash-out credit
card and store
3.Densest subgraph detection
11. Build Graph Model
Table: Notations and symbols
Symbols Interpretation
Tripartite graph of transfers in the bank
Set of nodes, indicating accounts
Set of edges, indicating transaction
Set of credit cards
Set of POS machine settlement account
Set of transaction account
Set of final debit card account where
fund flow trans to
Transaction between credit card and
merchant using POS machine
Transaction among different accounts
Credit card
purchase transaction
POS Machine Debit Card
13. Start from the whole graph, each time we can remove the node that has smallest contribution
to dense subgraph, in order to keep the node that can make density function approximates.
Build a priority tree , keep the weight of all its connected nodes and then choose the
minimum value of the sub node as the branch node.
Search dense subgraph
Greedy Approximation Search
Heuristic Optimization
Computational Efficiency
14. Result Disposal
Merchant closure rate 35%,
Disposal rate 75%.
Implement
Output top ten suspicion
subgraphs of GSQL
Result Based on Densest Subgraph Detection Display
Preparation
Credit card transaction data
and Settlement account
transfer data of all
merchants in the bank