New Digital World Opens up More Opportunities, Both for Good and Bad
The digital world has opened up endless possibilities for every aspect of our lives. It has enabled unprecedented behaviors such as virtual catch-ups with family and friends, remote work, and online shopping.
However, good things usually come with some sort of catch or a downside. Ironically, as technology continues enriching our lives by utilizing information, it has also made fraud more sophisticated. Mobile payment, e-commerce, digital banking, etc., traps are everywhere.
According to Consumer Sentinel Network Data Book published in February 2021, the number of fraud, identity theft and other illegal activities reports hit over 4.7 million in the United States of America in 2020.
Among all, 2.2 million was fraud. The total fraud losses were $3.3 billion, with a median of $311 losses. The most common fraud was imposter scams, where scammers ask people to send money typically by emails or calls or even on social media. Identity theft was the second most common with 1.4 million cases, targeting largely government documents/benefits, as well as credit cards and loans. Shockingly, the number of identity theft doubled from the previous year.
How Does Data Science Detect Fraud?
In the effort of flattening the increasing number of frauds, data analytics is becoming more prominent in the area of fraud detection across a wide range of industries. Our previous article “Top Insurtech Companies that Are Reshaping Insurance” briefly mentioned about insuretech finding its way as anti-fraud measures in the insurance industry. Likewise, other industries such as banking, financial services or telecommunication industries are increasingly implementing data analysis techniques to reduce fraud.
Knowledge Discovery in Database (KDD), Data Mining, Machine Learning, and Statistics are among these anti-fraud analysis methods. These techniques are classify into the two categories: statistical techniques and artificial intelligence. Some of the common statistical techniques are data processing that detects, validates, correct, and filling up missing or incorrect data, or regression analysis that estimates relationships between two or more variables.
These techniques become more sophisticated when combined with artificial intelligence, such as machine learning and data mining. Artificial intelligence is able to process larger datasets in less time than humans, creating algorithms to detect fraud more efficiently.
One of the simplest examples of the use of artificial intelligence in fraud prevention is spam detection in your email inbox. Just a quick look at your junk folder will tell you how many spams artificial intelligence has weeded out. Similarly, data science is protecting more and more businesses and consumers from con artists. The following section goes through some high-level examples of how data science is helping fraud detection in different industries.
According to the Data Book, credit cards were the most common payment method in fraud reports. The two most common credit card frauds involve:
- card theft where card details are stolen and used to make purchases.
- application fraud where false information is provided to obtain credit cards with the intention of not repaying.
With the use of data analytics, there are a number of ways by which possible fraud can be detected at an early stage. As technology evolves, better models and more data sources are emerging. Common strategies may include but not limited to:
- user-profile analytics, i.e., making risk profiles of high to low risk users, analyzing characteristics of the user such as age, income, location etc.
- transaction level analytics, i.e., reviewing patterns of transactions such as sudden unusual spikes in the amount or number, especially with new merchants/vendors, or change in times.
The Data Book shows insurance-related identity theft reports jumped by 62% in 2020 from the previous year. The common insurance fraud may include exaggeration of otherwise legitimate claims and intentional misinterpretation of the facts or claim processes. Professional fraudsters or by organized criminal gangs may commit these frauds, which makes it even harder to detect. Common fraud settings can be:
- over-invoicing claims, where individuals pair up with vendors (e.g. car mechanics) to inflate the repair costs to claim more.
- staged claims, where individuals stage fires or burglaries to make large claims to the insurance company.
In addition to user profiling mentioned in the credit card section, the following analysis may take place:
- vendor analysis, i.e., profiling and categorizing claims into homogeneous groups in order to find any outlier vendors with high claim costs and frequencies.
- structured learning, i.e., analyzing historical record of staged claims in order to identify the similarities among them and create an alarm that gives an early warning signal in case of a suspicious staged claim.
Phone call and texts are classic but remain as the most common contact method in fraud today, accounting for 31% and 27% of the overall fraud reports. Such telecommunication frauds may include:
- subscription fraud, where fraudsters obtain a subscription to a service with false identity details with no intention of repaying.
- superimposed fraud, where fraudsters use service without having the necessary authority, for example, to make calls using someone else’s subscription.
Since the telecommunication industry deals with a sheer amount of data, it faces different obstacles in detecting fraud. Examples of typical approaches include:
- outlier detection, i.e., using rules to detect outliers. For example, highlighting calls where the same number is used in two very distant locations.
- statistical summaries, i.e., analyzing statistical summaries of call distributions, comparing against ranges that are either determined by specialists or by application of supervised learning methods to the past fraud cases.
Data Science Brings More Efficiency and Accuracy to Fraud Detection
Data analytics with the use of artificial intelligence and machine learning tools has a great potential to improve business performance. Thereby, it can elevate customers experiences. Although there is still a need for reviews by human, artificial intelligence can exceed human precision sooner or later. One of the strengths of artificial intelligence is the fast-evolving ability to optimize and automate processes. Further development in data science could one day bring an end to the cat-and-mouse game of fraud.