Ai Editorial: Finding the right blend of data and machine learning for combating fraud

First Published on 13th February, 2017

Ai Editorial: The quality of data as well as making the most of different types of machine learning are vital for fraud prevention, writes Ai’s Ritesh Gupta

 

Fraud prevention isn’t just about one algorithm being used or acting only on historical data. One can fall woefully short with an ill-conceived approach. Airlines need to check valuable pointers – are chargeback rates under control? Even if the fraud system is indicating very low fraud rate, is it still resulting in high abandonment and rejection rates of the users?

Airlines are acknowledging the limitations of traditional rule-based fraud solutions, one of them being overly focused on bringing down the fraud rate as close to zero as possible. This tends to be a risk-averse approach, and one needs to negate rules when positive behaviour is detected. So how can big data and machine learning contribute?

Here we assess some of the critical aspects related with data strategy and machine learning that can contribute in fraud prevention:

·          Only predictive analytics isn’t enough: Predicting future fraud based on historical data isn’t enough. For instance, when transactions with no historical data are submitted into the system, the possibility of missing out on suspicious behavior is there.  

Unsupervised machine learning manages to seek patterns and correlation amidst the new data collected, which helps to identify positive and negative behaviour.

With pattern recognition, even without any prior historical data, the machine is able to discover patterns across various transactions and establish if the transaction showed bot behaviour or human behaviour. Information collected from big data is vital here. It is initially used to garner information about the user’s behaviour on the website and these details are blended with machine learning, which uses pattern recognition to chart the pattern of this user’s behaviour to match it either with positive (genuine) or negative (fraudulent) behaviour, as well as predictive analytics that records the positive/ negative behaviour and uses that on future transactions for potential signs of fraud. Also, behavioral analysis is one area that is becoming increasingly sophisticated. Swipes, taps, cursor movements etc. are being analyzed for navigation flow, time spent etc. to understand the behavior. Specialists are tracking mouse movements and clicks in context and meaning while becoming increasingly more accurate over time.  

·          Relevant data: While data is important, what is more important is the quality and relevance of the data. Big data is receiving greater popularity and used more widely than ever, but it is not about how big the data is. Relevant data is necessary to improve fraud prevention, as well as to improve the machine. For instance, if the machine is regularly receiving non-relevant data, the resultant output will be non-relevant decisions.

In addition, the way the data is processed must also be relevant when making probabilities of fraud risk.  Algorithms are designed with biases. If the fraud system’s algorithm is centred towards eliminating fraud entirely, the decisions will compile results of a very low fraud rate, but also high abandonment and rejection rates of the users. Instead, if the fraud system is focused on maximising revenue per risk of fraud, it is possible that a slight allowance of letting the fraud rates up by 0.1% could increase acceptance rates by 10%.

·          Keeping pace with technological developments: Airlines, just like any other e-commerce business, need to cater to a variety of payment methods, currencies and devices. Each new technological development introduces new venues for this fraud, meaning detection and prevention efforts need to be just as agile.  Expect more creative modes of fraud to appear. So there is a need to shift to active surveillance by deploying big data user and entity analytics to understand the user behaviour behind each transaction. Considering that most fraud attacks come as a coordinated attempt from a single script, automated to maximize the number of hits in the least amount of time possible, they will leave behind a pattern that can only be detected by understanding user behaviour. Even as new forms of payments become popular and mainstream, active surveillance will be more relevant (rather than static defence) and effective in dealing with fraudsters.

·          Airline-centric approach: As we highlighted in one of our recent articles, the e-commerce set up of airlines is distinctive, and it would be highly desirable to have a tailored data strategy.

For instance, how to capitalize on custom data fields be it for flight details, loyalty miles claims (to detect abnormalities) etc.

With more and more data analysed, it is harder for hackers to hide their tracks fully to pass off as genuine.

·          Assess liability shift: Airlines must expect better results at this juncture, considering that machine learning has evolved. For instance, improving fraud management doesn’t only mean lowering fraud rates, but it also about ensuring that the system does not hinder revenue growth. So better technology (big data, machine learning) is important, but how these systems are designed, and the KPI it keeps to is more important.

If we talk of liability shift, specialists point out that with pattern recognition, deep learning and stochastic optimization – seek an optimized yes or no decision in real time.

Based on calculated risks, the system passes the optimized number of transactions while ensuring that chargeback rates are still under control. As a result, borderline genuine transactions can be passed and unnecessary rules and bans are lifted, improving sales greatly. The efficacy of these “calculated risks” need to be scrutinized by airlines. It is being asserted that switch from predictive models of machine learning towards real-time machine learning is few years away.

 

Are you bold enough to survive in the brave new world?  Assess your preparedness at 11th Airline & Travel Payments Summit (ATPS).

Date: 03 May 2017 - 05 May 2017   

Location: Berlin, Germany 

For information, click here

 

Follow Ai on Twitter: @Ai_Connects_Us

Editorials

  • Ai Editorial: New weapons for combating false positives – are airlines ready? +

    First Published on 29th March, 2017 Ai Editorial: False positives have never been easy to identify. But with machine learning, 3D Secure 2.0 and data intelligence, airlines can be in Read More
  • flyiin’s Stéphane Pingaud on disruption via API aggregation +

    First Published on 27th March, 2017 flyiin is 100% API-based online marketplace that promises to show travellers offers that are relevant to them, and in doing so provides airlines entire Read More
  • Ai Editorial: Seamlessness and relevance - what every interaction should reflect +

    First published on 23rd March, 2017 Ai Editorial: Airlines can attempt to serve passengers in two ways – pre-empt what a passenger might opt for or being ready for the Read More
  • 1
  • 2
  • 3
  • 4
  • 5