Keyboard Dynamics

Mitchell A. Thornton Department of Computer Science and Engineering Department of Electrical Engineering Southern Methodist University

Synonyms dynamics, keystroke dynamics, typing patterns

Related Concepts and Keywords Bioinformatics; Behavioral bioinformatics, , handwriting and signature recognition

Definition Keyboard dynamics are the characteristics of typing or rhythmic patterns of sequences of keystrokes on a keyboard.

Background Early telegraph operators realized that they could identify other operators through the unique characteristics of their keying patterns. This differentiation was apparent although all operators were using the same Morse code form of communication since each individual operator develops unique keying characteristics. This same phenomena occurs in the patterns of keystroke sequences of the ubiquitous keyboard. Keyboard users develop and deterministically repeat characteristic patterns that exhibit some degree of uniqueness when compared to other users patterns.

Studies on the analysis of user rhythms exhibited by users of data entry devices occurred as early as the work of Bryan and Harter in 1895 when such patterns by telegraph operators were analyzed [1]. In terms of modern keyboards, an early analysis was published by Young in 1989 [2]. Past recent research is described in the work [3,4].

Theory

The rhythmic typing patterns of different individuals have characteristics that are unique. Because typing requires an interaction of the mind and manual dexterity, many complicated physiological processes are involved in the chain of events beginning with having a thought about which specific key on a keyboard is to be depressed (and released) and the actual event of key entry. This complicated chain of events is characterized by many biometric factors, or unique characteristics, attributed to a particular individual. Hence, these characteristics allow a unique “signature” to be measured based on the particular rhythmic sequence of typing a set of characters on a keyboard. This particular unique rhythmic sequence can be exploited or used in much the same way that a written signature is used for .

Two types of information are transmitted to a data processing device for each keypress, the actual character corresponding to the key and the duration of time that the key is depressed. Modern computers use the amount of time the key is pressed to determine if the user desires multiple instances of the character to be received. The interval in which characters are repeated is referred to as the typematic rate and is typically a variable parameter. Keystroke dynamic algorithms make use of the keypress duration at a more fine- grained level, however mechanisms are clearly in place to determine keypress durations. In addition to using the time between a single key down and key up event, a keystroke dynamic algorithm may also utilize the time interval between a key up event and a subsequent key down event of a different keystroke. The combination of these event time intervals then forms a signature. This signature is then compared to a predetermined and stored keystroke signature known as the baseline signature for the purposes of authentication.

Although the rhythmic sequence is unique, there are variations among subsequent key sequence entries. These variations can be characterized as short term and long term. Short term variations arise from factors such as an individual’s mood, state of alertness, or temporary physical injuries. An analogy with a written signature is useful in that an individual’s signature, when written twice subsequently, appears to be substantially the same but has slight variations. Such slight variations are also present in the rhythmic intervals between key depressions and, in fact, their variability is also a biometric attribute that varies among individuals. Measures of this variability over many subsequent training samples can result in the calculation of individual characteristic standard deviation values or other statistics.

Long term variations differ in that they represent gradually changing permanent characteristics. From a statistical point of view, these variations could be quantified as slowly changing mean values of individual key depression timing intervals. Long term variations typically occur due to a user becoming more adept or proficient in keyboarding skills through training, or through gaining familiarity with a particular keyboard and the nuances of its key depression pressures and spacing between the keys. The analogy of a handwritten signature to illustrate this situation is to compare an individual’s signature over the span of a lifetime. The signature would typically show distinct differences when compared at the time the writer was very young, then middle-aged, although these differences were acquired gradually over time.

A third type of variation that must be accounted for is an abrupt, but permanent change in an individual’s keyboarding characteristics. This phenomenon can arise due to circumstances such as a permanent injury to the hand or brain, or through acquiring new keyboarding techniques by taking a class to retrain the physiological chain of events leading to the depression of a key.

Many biometric methods are based on physical characteristics of an individual such as the pattern of a or iris. This method is actually a combination of characteristics and an interesting aspect of keyboard dynamics is that it actually incorporates the workings of an individual’s mind into the biometric. Other forms of biometric methods that include workings of the mind are facial gestures and walking gait patterns. Because the mind and the acumen of physical dexterity are adaptable characteristics, effective application of the method necessarily involves methods that account for adaptability. Furthermore, variability in the manufacturers of different models of keyboards also adds a degree of uncertainty. As an example, one keyboard may have pushbutton keys with slightly more or less pressure required to completely depress a key. This in turn can affect the resulting rhythmic typing patterns when the same individual uses two different keyboards and these differences must also be accounted for.

Applications The primary application of keystroke dynamics is its use in user authentication. The predominant features required in this application are to somehow account for short term, long term, and abrupt changes in a particular users’ characteristic keyboard dynamics.

While the application of keyboard dynamics can be used to authenticate a user at the initial stage of gaining access for a data entry session in a static manner, for instance during a password entry phase, an alternative application could be more dynamic in that keyboard dynamic metrics are continuously evaluated during the entire data entry session. Furthermore, the authentication could be accomplished in a background process by attempting to identify a user of a system that does not have a designated authentication stage. An example may be the use of a public data entry keyboard where passwords are not requested.

Another interesting application of keyboard dynamics is to utilize the signature to identify and couple a particular user with other known information about that user or group of users. For example, demographic profiles could be stored and accessed based on some category of users. Similar methods are commonly used in commercial webpages such as those employed by vendors of books. Once a particular user is identified, interest preferences that have been previously acquired are accessed and allow the webpage to customize a set of suggested products the user may be particularly interested in.

Experimental Results Several different studies and experiments have been performed in the past and are documented in the literature. The work described in [5] analyzed the different keystroke dynamic characteristics obtained when a group of users entered different kinds of text. The first kind was a sample of grammatically correct sentences that form a coherent passage in the English language, the second kind consisted of a collection of correctly spelled English words that were arranged in a random fashion, and the third type was a collection of random character sequences.

Another experimental result is presented in [6] where a study was made of grammatically correct English sentences versus users’ names and other words that they type very frequently. The result of this work allowed comparisons and analyses to be performed with respect to well trained or frequently used character string characteristics.

Open problems and future directions Because the use of keyboard dynamics has some variation due to the reasons discussed above, authentication errors have a non-zero probability of occurrence. These errors can be classified into two categories: a false positive where a particular user is incorrectly classified as the wrong person and gains access to a data entry system when they should not have, or a false negative, where a particular user should have gained access but was denied. Depending upon the application, one of these types of errors may be deemed more severe than the other. More research is required to allow for keyboard dynamic variability to be coped with while also reducing the probability of these types of errors. Current research is focused on the types and use of particular statistics to be employed.

To implement a keyboard dynamic approach, some initial characterization must be accomplished to establish the baseline signature of an individual or particular group of individuals. Depending upon the types of statistics employed, establishment of the baseline signature may require an unacceptably long or involved training period. Reducing the time and effort expended in establishment of the baseline keyboard dynamic signature while capturing an accurate representation is an area where further research is required. The use of areas such as machine learning algorithms, statistical identification theory, data mining, system identification, and control theory all have potential for application to the challenges associated with keyboard dynamics and are currently open areas.

The storage and use of the baseline signature is another area requiring careful consideration. Unintentional or malicious access of the baseline signatures could allow unauthorized users to gain access. The question arises and must be addressed concerning where the baseline signatures are to be stored within the system. If storage is implemented in the actual keyboard, the device could be easily accessed and the signatures obtained through reverse engineering. Alternatively, if the signature baseline data is stored in a remote secure server, some form of network security such as the use of encryption must be employed.

While the majority of past work has focused on timing characteristics in keyboard dynamics, some researchers have investigated the use of alternative metrics. Such metrics could be obtained through including other sensors in the keyboard that measure quantities such as the pressure profile, the position on the actual key where the finger usually makes physical contact, or the acceleration profiles that occur during the depression of a single key [7]. Incorporation of other metrics implies the need for a data fusion method to be accomplished to effectively combine the metrics into a single signature profile.

Determining if keyboard dynamics can be used to classify particular groups of users based on demographic information is also an interesting open problem. For example, native speakers of the English language are accustomed to typing certain alphabetic characters more frequently than others. This could potentially be exploited to identify not a single user, but the demographic group characterized as native English speakers. Other demographics that may be of interest could be based on gender or age. Because there is a strong analogy between keyboard dynamics and handwritten text, many of the ideas used in handwriting analysis could be applicable to keyboard dynamics. Handwriting analysis is a topic of research in the areas of human psychology and law enforcement.

Recommended reading

[1] Bryan, W.L. and Harter, N., Studies on the telegraphic language: The acquisition of a hierarchy of habits. Psychological Review 6, 4 (1989), pp. 345-375.

[2] Young, J.R. and Hammon, R.W., Method and apparatus for verifying an individual’s identity, U.S. Patent 4,805,222, February 14, 1989.

[3] Spillane, R., Keyboard apparatus for personal identification, Technical report, IBM Technical Disclosure Bulletin, 1975.

[4] Sternberg, S., Monsell, S., Knoll, R., and Wright, C. The latency and duration of rapid movement sequences: Comparisons of speech and typing, in G.E. Stelmach (ed.), Information Processing in Motor Control and Learning, New York: Academic Press, 1978, pp. 117-152.

[5] Gaines, R., Lisowski, W., Press, S., and Shapiro, N. Authentication by keystroke timing: Some preliminary results, Rand Report R-256-NSF, Rand Corporation, 1980.

[6] Bleha, S., Slivinsky, C., and Hussien, B. Computer-access security systems using keystroke dynamics. IEEE Transactions on Pattern Analysis and Machine Intelligence, December 1990, vol. 12, no. 12, pp. 1217-1222.

[7] Allen, J.D. and Howard, J., Design and implementation of a novel behavioral biometric for user authentication. Proceedings of the Society for Design and Process Science, 2010.