Fighting Fraud with Machine Learning

Table of Contents

  1. Introduction
  2. Entities fighting fraud
    1. Government fighting fraud
    2. Financial Institutions
    3. Tech Companies
  3. Ways of fighting fraud with machine learning
    1. Clustering Methods
  4. Products
    1. Amazon Fraud Detector
    2. Feedzai
    3. DataVisor
  5. Concluding Remarks


  1. Introduction

Fraud detection is an extremely costly industry. Fraud is rampant in the United States and across the world. The cybersecurity firm, McAffee, estimates that cybercrime costs the world $600B. There were 3 million identity theft and fraud related reports submitted in 2018 in just the United States. Companies fight fraud daily to prevent losses before they happen. However, in the digital world, software is needed to identify fraud as quickly as possible before losses mount.

More recently, machine learning has been employed to identify and fight fraud. Companies such as MasterCard are leading the charge and are not sitting back to analyze transactions after they’ve occurred – the name of the game is in real-time detection.

  1. Entities fighting fraud

Fraud spans nearly every industry. Humans are always looking for ways to cheat the system and find loopholes for their personal game. As a result, entire industries have their own methods of fighting fraud.

  • Government fighting fraud

Many government institutions are turning to software to help in the fraud battle too. Among them is the Federal Reserve which is looking to limit the amount of misreporting that occurs. There is more than $140B a year that is wasted from the federal government due to fraud or abuse as a result of government programs according to a McKinsey Company study. Some of this is due to misuse of funds that are distributed from these programs. The government is turning to increased efforts with data analytics and machine learning to combat these losses.

  • Financial Institutions

Financial institutions are perhaps the first entities that come to mind when you think of fraud. Banks take the brunt of fighting against fraud since the money needs to come in and go out. Financial institutions such as Chase use machine learning to protect customer data. A interview with Chase executive, Andrew Sloper, elucidates how large financial institutions have multi-faceted approaches to security. They research the latest instances of fraud to stay up to date as well as both supervised and unsupervised machine learning.

West Monroe Partners surveyed 150 executives at medium-sized banks across the U.S. and found that 61% of them are turning to automation including machine learning in order to increase efficiency. That number is rather low compared to other industries as stated in the article. However, all banks will soon need to adapt and become more accepting of automation and machine learning approaches.

  1. Ways of fighting fraud with machine learning

Methods of fighting fraud span many different types of machine learning techniques depending on the industry as well as the type of problem that is being solved.

  • Clustering Methods

Certain methods are fantastic tools to use against fraud, such as clustering analysis. These types of visualizations such as principal component analysis allow users to see high dimensional data in something more palatable.

A great interview with one of the executives at SAS, Stu Bradley, found here discusses some different approaches that companies can take to fight fraud. In the article, Mr. Bradley states how a common approach is to give an account a primary fraud score and then drilling down deeper to determine the likelihood of fraud occurring at the transaction level.

  1. Products

There are several leaders that create products to fight fraud with machine learning techniques. These solutions are purchased by some of the largest organizations in the world to assist with loss mitigation as well as fast detection of wrongdoing.

  • Amazon Fraud Detector

Mainstream companies are developing solutions to help fight fraud. One of the most familiar with this problem is Amazon who is the world’s largest ecommerce retailer.

Recently at their AWS ReInvent conference in Las Vegas, AWS released Amazon Fraud Detector. The detector can look for anomalies in transaction data. Some example data that could be provided to the fraud detector would be IP addresses in addition to historical transaction data. AWS can then sift through the data to identify malicious actors and then create rules to limit the actors’ interaction with the customer’s platform.

The financial crime fighter, Feedzai, claims that they can prevent payment fraud before it happens.


Some of their metrics are that they have been able to detect 82% of fraudulent cards at a leading credit card issuer. They allow integration with many of the current state-of-the-art systems such as DataRobot as well as Python and R.

  • DataVisor

Another solution that is available in order to fight online fraud is Datavisor. The company touts 94% detection accuracy along with just a 0.17% false positive rate. False positive rate is a metric that is very important with these types of solutions since customers hate when their card is flagged or declined when it is not due to a fraudulent transaction.

One case study on the Datavisor website covers a unicorn food delivery service. The company had over 100M users and had a product where users could search, find, order, and have food delivered to them.

The solution that DataVisor ended up implementing was using an unsupervised machine learning model to find bots that registered accounts. They claim that the company was saved $6M in fraud losses.

  • Fico Falcon-X

Several of the features that Falcon-X features are real-time anti-money laundering measures as well as real-time decision-making suites powered by strong cloud-based compute. This may be one of the most battle-tested solutions out there with the Chief Analytics Officer, Scott Zoldi, stating that the Falcon software suite has been worked on for the last 25 years through numerous iterations. The software holds 95 patents and currently protects 2.6B accounts worldwide.

  1. Concluding Remarks

Fraud will always be an on-going problem that organizations must face. Thankfully in the world of machine learning, some of the heavy lifting can be done for us. Models can be trained that can identify fraudulent transactions with low false positive rates automatically so that the attacks are found immediately when they happen. Companies such as Amazon are even working on solutions to these types of problems.




Show More

Nick Allyn

Hello, my name is Nick Allyn. I am extremely passionate about the field of artificial intelligence. I believe that artificial intelligence will save millions of lives in the coming years due in higher cancer survival rates, cleaner air, as well as autonomous cars.

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button