First Published on 14th June, 2018
Ai Editorial: Deploying a multi-disciplinary approach combining different technologies - both supervised and unsupervised machine learning (ML) - would better equip merchants to deal with fraud management, writes Ai's Ritesh Gupta
The travel industry needs to dig deeper to understand the efficacy of machine learning and its role in curbing payment fraud as well as the rising issue of account takeovers.
Machine learning often encompasses different types, and simply using one type (predictive analytics) is insufficient.
Supervised machine learning is considered to be a reactive approach to treat fraud. It has contributed in combating fraud to a certain extent – automating some processes, garnering more data to evaluate, but the industry has to capitalize on real-time machine learning as well.
Without real-time learning, supervised machine learning is unable to forecast and offset unfamiliar fraud attacks, since it is dependent only on the data on previous fraud attacks. Also, these systems can only generate probability scores for each transaction, therefore still involving manual reviews.
Many fraud solutions on the market are built with machine learning, but they are built with only one machine learning model (e.g. Random Forest) and the belief that relying on one model will be sufficient in allowing them to detect and prevent coordinated fraud attacks, says Justin Lie, CashShield’s CEO.
"Most travel e-commerce merchants still rely on this single disciplinary approach, requiring historical data to make correlations detect anomalies. However, as fraudsters become increasingly sophisticated, using machine learning for their attacks, they can get ahead by flooding systems with so much fake data that they pass through undetected," cautioned Lie.
Lie added, "As such, deploying a multi-disciplinary approach combining different technologies - both supervised and unsupervised machine learning - would better equip merchants to deal with fraud management. Unsupervised machine learning can be used to learn on the fly and identify fraudulent patterns even without having been trained with historical data, i.e. able to identify unknown fraud attacks. Thereafter, predictive analytics may still be used to run the probabilities of fraud, giving a risk score."
Unsupervised machine learning is able to seek patterns and correlation amidst the new data collected, which helps to identify positive and negative behaviour, and is effective in identifying genuine customers as much as identifying fraudsters. Specialists recommend that pattern recognition, deep learning and stochastic optimization are also necessary for an optimized yes or no decision in real-time.
Making it work
Lie explained how the combination of unsupervised machine learning and supervised machine learning can work best in curbing fraud. He mentioned:
Blend of big data and machine learning
The combination of big data and machine learning allows more effective fraud prevention. Big data is first used to garner details about the user’s behaviour on the website (for e. g. the movement of the mouse) which is combined with machine learning. There is use of pattern recognition to configure this user's behaviour to tally it either with authentic or fraudulent behaviour. Along with this predictive analytics comes into play to record the positive/ negative behaviour and avail that on future transactions for probable signs of fraud. Finally, an optimized fraud risk algorithm needs to be counted upon to make decisions on whether or not to accept a transaction based on calculated risks to best optimize sales while controlling fraud and chargeback rates.
"Big data allows for more data collected - but relevant data is more important than collecting more data. Collecting data from the merchant’s website and behavioral data beyond payment data will be useful for analysis on the user’s behavior - whether good or bad," mentioned Lie.
A transaction may be sliced into multiple data points, where it may then be combined with real-time machine learning to match patterns through the permutations and combinations of the data points, as well as to identify when fraudsters make micro-changes between transactions (such as changing the device from iOS to Android between transactions to seem like the transactions come from a different source). As it turns out, most systems are still relying on a single disciplinary approach, and a multi-disciplinary approach that combines big data, predictive analytics and real-time machine learning would be more effective in detecting coordinated fraud attacks, recommended Lie.
Act and take charge
Travel merchants need to defend themselves adequately by using machine learning, and at the same time there needs to be reliance on rules and the human component (intervention and feedback) as well.
Merchants should learn to discern and understand the different types of machine learning, and be sure to know if the fraud solution uses only predictive analytics or covers more bases with more than one kind of machine learning. Machine learning technologies are yet to be deployed commonplace to secure accounts, even though machine learning, especially real-time machine learning can be applied on account protection.
Lie concluded with a word of caution for merchants: Many merchants are also still reliant on manual reviews, which means that even if they were able to improve their machine learning algorithms and systems, they would always still be held back by the end process of manual reviews and human errors.
Hear from airlines and other industry executives about travel fraud at the upcoming 7th Annual Airline & Travel Payments Summit (ATPS), co-hosted with UATP, (4- 6 September 2018 in Phuket, Thailand).
For more click here
Follow Ai on Twitter: @Ai_Connects_Us