The risk of stealing credit cards has become one of the problems that plague the global bank credit card department. In the United States alone, the Federal Reserve’s payment survey shows that the total amount of credit card payments in the United States reached $26 billion in 2012, of which unauthorised credit card payments, or stolen credit cards, amounted to $6.1 billion.
For banks, measuring the risk of credit card transactions is very difficult. To achieve this goal, you must quickly determine which transactions are legally authorized and which are stolen. So how are these tasks achieved?
From the standpoint of consumers, the process of detecting credit card theft seems to be "magic". This kind of detection is almost instantaneous, and involves a series of complicated technologies, from finance to economy to law to information science. Of course, some credit card theft detection is very simple, for example, when the machine finds that the zip code of the credit card is not in accordance with the zip code of the issuing place, it will issue a warning.
Traditional credit card detection requires a lot of manpower to participate in the analysis and judgment of massive data. The algorithm only warns a transaction, and finally the human auditor will call to confirm whether the transaction is suspected of credit card theft. Now, due to the surge in transaction volume, the credit card departments of major banks have begun to rely on big data and quickly identify unauthorized credit card transactions through machine learning and cloud computing.
The machine learning algorithm used for credit card detection will first be trained by a large amount of normal transaction data and cardholder data. The result of the transaction will become an important dimension of the machine understanding of the transaction, such as a normal person may buy electricity once a week, go to a shopping center every two weeks, etc., these transaction results will become a model of normal transactions.
Next, the machine will accept the test of real-time transaction data and give the probability that the transaction is illegal, such as 97%. If the detection system sets the probability that each transaction is illegal, it cannot be higher than 95%, then all these transactions will be Refusal to accept payment, in other words, the transaction will not succeed.
This algorithm considers many factors, including the trust of the card supplier, the card purchase behavior (time and space dimension), IP address, etc. The more factors are considered, the more accurate the model is built.
The process of this test is almost real-time, which is also the speed of detection that human workers cannot achieve. But the entire process still requires human involvement, including human review of algorithmic judgments and subsequent credit card fraud tracking.
The data in the financial transaction process is very large, for example, PayPal currently handles 1.1 PB of data for 16 million users. But for machine learning, the more data it means, the better it can improve its algorithmic accuracy, so that it can better identify theft of credit card events. Massive data has very high hardware requirements for bank IT systems, and data storage, reading and analysis are huge IT expenses.
At present, the credit card department of the bank has begun to use cloud computing as a way to process massive data. The flexible and scalable nature of cloud computing enables the efficiency of machine learning algorithms to effectively cope with credit card detection during peak hours such as double 11 card swiping.
The war around credit card theft and anti-piracy has continued, and with the help of machine learning, big data and cloud computing, and blockchain technology that can be implemented in the future, our wallets and card packs will become more and more secure in the future.
Heated Mouse Mat,Heated Computer Mouse Mat,Usb Heated Mouse Mat,Heated Mouse Pad Warmer
Ningbo Sinco Industrial & Trading Co., Ltd. , https://www.newsinco.com