Artificial Immune System for Fraud Detection

Download Report

Transcript Artificial Immune System for Fraud Detection

By : Anas Assiri






Introduction
fraud detection
Immune system
Artificial immune system (AIS)
AISFD
Clonal selection
Number of
Fraud cases
Credit card
transactions
Need for
Fraud
detection
• deviation detection
• outlier analysis
Fraud
• anomaly detection
detection
• exception mining
• mining rare classes
The fraud detection system with the increasing
transactions should have more characteristics
Diversification
Selfadaptation



The immune system is highly complicated
detecting and eliminating infections
similar problem: credit card fraud detection



Artificial immune system (AIS)
Immunological computation
is an area of biologically inspired computation.
Anomaly
detection
Patten
recognition
Data
mining
computer
security
adaptive
control
fault
detection

AIS and other soft computing paradigms:
 ANN (artificial neural network)
 EA (evolutionary algorithms)
 FS (fuzzy system)
 PR (probabilistic reasoning).
Since only one
financial transaction
of a thousand is
invalid
Credit card fraud
is rare event
related to the
total credit card
records.
AIS have the advantage of
distinguish between :
• non-self cells ( fraud attempts )
• self cells ( normal transactions ).



AISFD is a system based on AIS and casebased reasoning (CBR).
Artificial Immune system is inspired from
biological immune system.
Tow main algorithms :
 Negative selection
 Clonal selection

Negative selection
 The purpose of negative selection is to provide self-
tolerance to T-cells. It detects unknown antigens,
without reacting with the self cells.

Clonal selection
 The clonal selection principle describes the basic
features of an immune response to an antigenic
stimulus. It can make sure only the cells that can
recognize the antigen can proliferate.

Gene library evolution
 Gene library evolution learns knowledge of currently
existing antigens.


library of past cases
rather than being encoded in classical rules
Each case typically contains :
 a description of the problem
 and a solution and or the outcome

The knowledge and reasoning process used
by an expert to solve the problem is not
recorded, but is implicit in the solution
To solve a current problem:
the problem is
matched
similar cases are
retrieved
suggest a solution
problem and
solution are
retained as part of
a new case.
solution is revised
reuse and test for
success

Case library initialization (normal
transactions and fraud transactions case)
 With supervised classification

Antibody Gene (detectors.)library
Initialization
 Random or genetic algorithms


Negative selection of detectors
Clonal selection of detectors
Fraud detection
When a new transaction is submitted for fraud
detection, the fraud detection function is
activated. The affinity between the antibodies
(detectors) in the gene library and the new
antigens is calculated. If the affinity threshold
set by the system is exceeded, the AISFD
system sends out a fraud alert.







Diversification
Online incremental learning
Self-adaptation
Distribution
Fast response
Multi-level classification
Thanks