Researchers at the Technion-Israel Institute of Technology have developed software that is able to identify computer users by their individual, distinct typing styles. This ‘behaviometric’ technology may one-day be part of security systems to prevent unauthorised users from gaining access to computers and sensitive data.
According to a statement, the technology can be extended to other applications that involve a sufficiently complex interaction between humans and machines. Examples of such applications would include the identification of unauthorised drivers or pilots.
Students Ido Yariv and Mordechai Nisenson, under the supervision of Technion Professors Ran El-Yaniv and Ron Meir developed the system prototype in the Data Mining Lab of the Technion’s Computer Science Faculty.
‘This software is based upon a universal prediction algorithm,’ explained El-Yaniv. ‘It utilises statistics gathered while a person types freely, and learns the specific behaviour patterns that accurately identify the typist.’
El-Yaniv further explained that time differential patterns between consecutive keystrokes can uniquely determine a user. In some cases, this can be accomplished after only a very few keystrokes.
The system’s innovation is the ability to identify the legal user as he or she types freely, a stark contrast to currently existing technologies that can only achieve equally high recognition rates if both the legal user and the intruder are asked to type the same long sentence.
The system’s accuracy is said to depend upon the length of its training. But even after a relatively short session consisting of several typed sentences, the system can distinguish the user from potential intruders with around 90 percent accuracy from a sentence as short as ‘What did you do today?’
After its initial training, the system continues monitoring the user and obtaining more keystroke sequences, allowing it to reach extremely high rates of recognition.
According to El-Yaniv, typing is a complex sequence of events. Many processes and factors make up this sequence, including language fluency, mental attributes, fine motor skills, hand size and finger length.
As part of their ongoing research, the group will study the system’s reliability in the face of ‘noise’ that can be introduced by many factors such as a finger injury.
El-Yaniv believes that since it is the user’s behaviour that is being measured (through the timing of keystrokes), there would still be enough information by which to make a very accurate identification.