The ATMs They carry some danger. The thieves have learned to ‘hack’ them and cases such as the scanning of cards and keys or changing the card. An investigation of the National University of Colombia has devised an artificial intelligence system that helps predict ATM fraud.
For the study, the recordings of the security cameras of the ATMs were used, with which the neural network was fed so that it learned the movements, distances traveled and actions that the criminals were carrying out during the robberies.
What system is used to predict ATM fraud?
A artificial neural network –system of connections designed through programming and statistical modeling–, in this case convolutional and Siamese type, allows recognize the people and objects in the videos and compare all the data. With this system it was possible predict up to 86.7% of ATM fraud executed using the modus operandi of the exchange; in turn, it predicted with great precision when it is not, with a percentage close to 95%.
Gerardo García Arias, a student of the Master of Science with an emphasis in Statistics from the National University of Colombia (UNAL), compiled about six months of recordings from these cameras, in about 25 places in Bogotá, in which 43 frauds were witnessed. of the exchange mode. The data was entered into the network together with all the recordings in which a robbery was not recorded, which were around 300, so that the system could compare and recognize whether the movements were suspicious or not.
The researcher explains that a structured database was built based on the routes, which used a generalized linear model for binary response, which analyzes the information collected by the network and estimates the probability of occurrence of the fraud pattern by evaluating the distance traveled and the number of individuals, among others.
“The greater the distance traveled, the more likely it is to be a fraudand if we add to this the fact that there is a pattern of leaving and returning to the place, and the repeated presence of criminals around the ATM, the risk increases”, he assures.
What images have been used in this study
The videos are made up of images that go by very fast every secondand within them there are pixels, which are small information boxes that allow an image to be seen, in this case with the location of each user and the identification of their routes and movement patterns; Thus, when talking about distances in the videos, we are talking about the number of pixels, which were compared between the recordings of the different places.
It was determined that the suspicious distances were around 10,500 standardized pixels, which was contrasted with the average during a normal operation, which is 4,368. In addition, the returns to the place presented by each individual and the number of people who were there, and the fact of visiting two or more ATMs were analyzed.
“When people accept help from third parties, the loss of money is not the responsibility of the banks; however, for the financial institution it may entail a risk to its reputation. The statistics allow us to have a real option that would help to attack the problem, which would be by predicting through movements whether or not there may be fraud, protecting both the entity and its clients”, indicates the expert.
The “cambiazo” begins when the organized group blocks the ATMs, for example by putting adhesive tape on the point where the card is read. Then, when a person enters to make a withdrawal, which is unsuccessful because the ATM is altered, a second criminal enters who takes the opportunity to “help” him, obtaining the card and later his password, with which he can extract the money from the account , from another place.
Finally, the researcher stresses that these tools could also reduce the costs of manual and visual monitoring of security personnel hired to combat fraud, also reducing the places where this personnel is needed, since there would be automatic alerts that would allow that the actions are focused on the key points.