Another machine learning skill: detecting financial fraud
Such a thing happens every day all over the UK: a customer checks out with a credit card at a grocery store, but the cashier says that the card was rejected, but the things he bought have been packed, which is embarrassing. However, the credit limit of the card is good, and the PIN code is correct. Customers have bought many other things before, and now it is really baffling that the credit card cannot be used.
For the financial services industry, these "miscalculations" have become a growing problem. Not only do customers and merchants worry, but also staff are required to verify identity and unlock cards.
Every financial service company claims that it has a special fraud detection system that can distinguish between normal use and fraud, but it cannot guarantee that the cards of domestic users will not be rejected without reason. No matter how complex these systems claim to be, in fact, these false positives have indicated that there are problems within the system.
Cambridge Company
So, in order to solve the problem of the system platform, a company founded in Cambridge launched the adaptive behavior analysis technology, which can more accurately judge the quality of transactions. The company is FeatureSpace, and its platform uses ARIC engine and machine learning system to monitor whether the small details of complex events are abnormal.
Recently, Google's DeepMind, another Cambridge founded enterprise, has attracted worldwide attention. The company also uses the concept of machine learning and Bayesian statistics. This system is famous through robot and human champion Go competitions, but we have only seen a little of its power, because machines will infiltrate into the array of huge decision-making systems in the next decade.
FeatureSpace was a concept project of Cambridge University ten years ago, and was developed slowly and steadily by Professor David Excell (CTO of the company) and Professor Bill Fitzgerald (died in April 2014), which also proved that the company worked hard. With the appointment of a new CEO, Martina King, in 2012, the strength of the company seems to have been significantly strengthened. In 2014, investors formed a consortium and raised a total of 4.5 million dollars, including Mike Lynch, the co-founder of Autonomy, a famous Cambridge company.
During the conversation, Martina was obviously very busy (she had led Yahoo UK and Capital Radio in her long career). She was warm, considerate and likable. Engineers sometimes limit themselves to technical details, but Martina's vision of FeatureSpace is clear and concise.
"When you are doing something new and different, it will take some time to get people's recognition. Then they will know that there is a better way," Martina said. She firmly believed that although adaptive behavior analysis was once a profound laboratory technology, now everyone's views are beginning to change. " This is the result of word of mouth. "
Through Mike Lynch, Martina met Professor Fitzgerald, the co-founder of FeatureSpace. Although he passed away in 2014, she still misses professors and is very disappointed.
On the other hand, Martina is not complimentary about the fraud detection system used by the bank. " They [financial services] have lost so much money, "she sighed. "These attacks are so complicated that you can't predict their next step."
In the past few years, the problem of false and missed fraud and customer dissatisfaction has become an important issue from inconvenience. But why do traditional systems struggle? Martina thinks the reason is very simple. Their fraud detection rules are based on known model patterns. As soon as something new appears, it will lead to system crash or management becomes expensive. Therefore, the detection of fraud will eventually develop to expensive and manual processing.
ARIC simulates more detailed fraud in the real world and discovers anomalies by understanding the context of events. This is a theory to understand that the real world has gained traction in various forms of computer security, but it does not mean that its successful implementation is insignificant.
"It's hard to do this. You need to really understand statistical analysis and provide analysis in this regard within the enterprise," Martina said. "We have created markets in many ways. Until recently no one talked about machine learning."
Its strength lies in that all events, including fraud, are carried out by or on behalf of human beings. This gave Bayesian mathematical support, and ARIC not only grasped the meaning of an event in its search, but also predicted what might happen next.
The challenge faced by FeatureSpace is simply to detect fraud more cheaply and faster.
"When I wake up in the morning, how can I shorten the time?"
Martina estimates that the global cost of responding to false positives alone may reach $6.4 billion, which accounts for a large proportion of true fraud. If solving fraud becomes a heavy burden, then the problem needs to be solved urgently.
"We know that FeatureSpace can reduce alerts by 70%, so we don't need so many people to handle misjudgments," Martina said.
future
The company has more and more customers. Martina is particularly proud that it signed a five-year contract with Zapp, a mobile payment processor, in 2015. Zapp and Barclay worked together in the UK. FeatureSpace processes thousands of transactions per second, proving its technical depth and breadth.
ARIC is also used by a large American bank and a series of other financial institutions. However, Martina cannot disclose the company name. It is necessary to keep its products confidential for competitive reasons.
What ultimately determines the company's continuous growth is not simply its customer list, but the whole situation of anomaly detection itself. Following the road Autonomy opened up ten years ago, DeepMind is now world-famous, and FeatureSpace is also a combination of time, place and people.
At the critical moment, Cambridge has surpassed its reputation as a place of interesting computer science and smart mathematics, and established its position as one of the most important machine learning centers in the world. Martina believes that computers are well used not only to make the world work differently, but also for a better way. Otherwise, e-commerce will sink slowly because it cannot provide reliable transactions.
