Design & Implementation of a Fraud Detection System for Autonomous Teams - Kolloquium Louis Andrew - 570624 02.08.2022

Design & Implementation of a Fraud Detection System for Autonomous Teams

Louis Andrew - 570624

1. Supervisor: Prof. Dr. Christin Schmidt 2. Supervisor: MSc. Tobias Dumke 02.08.2022

Table of Content

  1. Motivation
  2. Goal
  3. Analysis
  4. Features
  5. Evaluation

Motivation

  • Fraud activities can damage a company's brand and profits
  • 70% increase in fraud activities in 2021, compared to 2020 [1]
  • Large businesses usually built upon multiple independent teams with different responsibility and expertise
Source: https://public.tableau.com/app/profile/federal.trade.commission/viz/TheBigViewAllSentinelReports/TrendsOverTime

Goal

  • Establish a common ground for collaboration between multiple teams in a single fraud detection process
  • Each team can contribute their expertise and domain knowledge in the process
  • Simplify a customer validation process
    • Encapsulating the internal logic
    • Presenting the validation result in an understandable format

State of the Art

  • Fraud detection techniques can be categorized to two categories [2]:
    • Supervised
    • Unsupervised
  • The intent of this research project is
    • Not to come up with a better way to predict a fraud
    • But to design a system that can facilitate collaboration between multiple teams in a fraud detection process
  • Rule-based approach is used as the fraud detection technique of choice in this project
    • Has some disadvantages: can be difficult to manage, rules have to be adapted
    • But is the simplest to implement and has a potential to enable collaboration

Use Cases

Use case diagram created to visualize the possible use cases of the system

User Stories

User stories are defined by reexamining the use case diagram. The user stories act as the main requirements of the system.
  • Verifying Customer Early “As a stakeholder, I want to verify customers, so that the company can have more confidence that the existing user base is trustworthy“
  • Notification on Suspicious Cases “As an employee, I want to be notified when a user seems suspicious, so that I can do necessary actions accordingly“
  • Validation Rules Management “As an employee, I want to manage my own rule to validate users, so that I can use my expertise to find suspicious customers as efficiently as possible without the communication overhead with other teams“

System Architecture

A system architecture diagram describes each component of the system and the interaction between each component.

Software Architecture

Software architecture diagram is created to better visualize the functions and role of each component.

Customer Validation on a Registration Event

“As a stakeholder, I want to verify customer, so that the company can have more confidence that the existing user base is trustworthy“ A possiblity to trigger a validation process after each registration should be implemented, so that the customers can be validated as soon as possible.

Notification on Suspicious Cases

“As an employee, I want to be notified when a user seems suspicious, so that I can do necessary actions accordingly“ The following sequence should be implemented as a way to notify concerned parties on certain suspicious cases:
  • After a validation process is completed, the system should publish a message to a messaging system
  • The messaging system should then forward the message to its consumers
  • The message consumers can run certain actions independently if needed (such as sending an email to alert a team member)

Managing Validation Rules

“As an employee, I want to manage my own rule to validate users, so that I can use my expertise to find suspicious customers as efficiently as possible without the communication overhead with other teams” The system facilitate collaboration by providing each team the possibility to create, read, update and delete their own validation rule.

Validation Real-Time Progress

Even though the user can receive a notification on certain cases, there might be times when a user wants to intentionally monitor the progress of a validation process. The sequence of such functionality will be as follows:
  • The system receives a new validation event and runs the validation process
  • The system provides a subscription endpoint that accepts a validation ID as the input
  • The system can establish a long-lived HTTP connection to the client through the subscription endpoint
  • Whenever a validation rule evaluation is completed, the system sends a message to the client

Calculating a Fraud Score

Evaluation

Title of User Story
Implementation
Verifying Customer Early
  • Provide HTTP endpoint to trigger a validation process
  • Returning a probability of the customer being a fraud at the end of a validation process
Notification on Suspicious Cases
  • Providing an AMQP message broker
  • Publish message to AMQP exchange whenever validation process is complete
Validation Rules Management
  • Provided endpoints to create, read, delete and update validation rules

Sources

[1]
New Data Shows FTC Received 2.8 Million Fraud Reports from Consumers in 2021. Online: https://www.ftc.gov/news-events/news/press-releases/2022/02/ new-data-shows-ftc-received-28-million-fraud-reports-consumers-2021- 0; latest access: 01 VII 22. Feb. 2022.
[2]
Richard Bolton and David Hand. “Unsupervised Profiling Methods for Fraud Detection”. In: Conference on Credit Scoring and Credit Control 7 (Sept. 2001).

Thank You