Technical References - [biotype]

BioType is a behavioral biometrics WebDNA function based on ADGS research and development (from version 8.5)

BioType is based on the concept of behavioral biometrics, meaning the way people do things individually, such as speaking, writing, typing, walking styles.

BioType is a WebDNA function: it is built-in

With BioType, WebDNA improves the way users are authenticated, ensuring imposters masquerading as the valid user are unable to access private or corporate information.

WebDNA's approach towards user validation and authentication incorporates a series of biometric security algorithms that record and measure how a person uniquely types their credentials. The multi-dimensional product combines the user's typing information with other behavioral and environmental heuristics to ensure that rightful users are granted access to the system. WebDNA technology not only automatically detects and prevents fraudulent authentication attempts, but it renders stolen credentials useless and limits invalid password sharing.

The beauty of WebDNA BioType's keystroke dynamics software lies in adding an extra layer of protection that is highly cost effective, accurate and user friendly. No additional hardware is needed, data are collected from users without being intrusive.

JavaScript is used to capture the keystroke from a web form that send the data to WebDNA. The BioType function builds a profile for the user and stores it into a database.

There are three methods:

- initialize records the keystroke dynamic for a specific user. He will have to type his credentials from 2 to 4 times to get enough data to build the profile.

- evaluate Once the profile is built and stored for a user, [evaluate] will compare every new credential typing and return a value from 0.1 to 4.0
0.1 to 0.5 is an evaluation that can be translated as "almost certainly the same user" (>90% chances)
0.5 to 0.8 is probably the same user (80% to 90%)
0.8 to 4.0 is most probably an impostor

- train may be used by the administrator from time to time to rebuild the profile, if the initial conditions change: different keyboard layout, health or physical problems etc...



By passing the user name and the method initialize along with the captured data,
WebDNA will create a profile for the user. The method evaluate can then return a deviation [btuser_deviation].

ParameterDescription
methodinitialize/train/evaluate
captureVariable for the KSD data from javascript
btuserName of the BioType user, ignored for demo
bttypeSet to "TEXT" for large blocks of text, otherwise defaults to password. This controls how BioType evaluates the keystrokes
btcorrectionsMaximum number of corrections allowed, defaults to 1
btlengthNumber of keystrokes for the text/password, defaults to 8
btthresholdThis parameter allows to specify the [user_deviation] value for which a user will be considered "impostor". Anything above it triggers the imposter. Below is "legitimate". The result is shown with [biotype]


When running "evaluate", if the result is "LEGITIMATE" then afterwards BioType will train with the data: this means that the data will automatically and transparently be used to train the system for this user. If the result is "IMPOSTER", then it will not train or modify the database for the user.



BioType with Free Text


A future version of WebDNA will also integrate a different keystroke dynamics service for free text. This method is “not what you type, but how you type.” Accurate recognition of free text keystroke dynamics is challenging due to the unstructured and sparse nature of the data and its underlying variability.

In this approach, the user types in text, as usual, without any kind of extra work to be done for authentication. Moreover, it only involves the user’s own keyboard and no other external hardware.

Free-text methodology does not restrict users to a particular text; on the contrary, a user is given complete freedom to use any text of any length without any constraints. BioType will continue to collect the keystrokes, after successfully passing the identification session, throughout the whole time that the user is logged-in. The user’s typing pattern is typically monitored during several days where he/she is performing regular typing tasks such as writing e-mails or typing word documents. The enrolment phase is transparent to the user.

Evaluation and Results of the Free Text BioType method, by ADGS :
The methods used by BioType for WebDNA are based on reasearch and developments from ADGS. A dataset containing more than 20,000 sessions from 20 different users was used to evaluate the performance of free text keystroke dynamics authentication method. The sessions were captured during a one year period to reflect long term variations in typing rhythms. The users moved between different computer terminals and keyboards and typed under varying conditions of tiredness and stress.

As usual with soft biometric authentication methods the error rates are not single fixed numbers but random variables, not only considering a set of users but even when considering a single user, because of the variations shown in biometric templates and sometimes in the classification methods.

The performance of the authentication method is measured using two error rate metrics: False Acceptance Rate (probability that an impostor is accepted as a legitimate user) and False Rejection Rate (probability that the legitimate user is rejected by the system)



An effective keystroke is an alphanumerical or space keystroke, thus excluding modifiers, functions, special and navigation keys. A session is a sequence of keystrokes typed by a single user in the normal course of his daily work, possibly including pauses and interruptions, both naturally occurring and external, including at least 100 effective keystrokes; the user might have left the computer terminal and resumed writing, as long as the interval in a single session remains under three minutes.

Fig 1

Fig 2

Fig 3

Fig 4

Fig 5

Fig 6



In figures 1 and 2 the evolution of FAR and FRR as a function of the number of available training sessions is shown; the plot depicts average, one standard deviation channels, maximum and minimum. Figures 3 and 4 show the error rate evolution for the best and worst users in the data, as a function of the number of training sessions and of the effective keystroke count. Figures 5 and 6 show the variation of the FAR and FRR ranges for a single user with a high variability in error rate if different training sessions are used.

The asymptotical average error rates are 0.92% for FAR (with a standard deviation of 0.74) and 4.28% for FRR (with a standard deviation of 2.73); both distributions are positively skewed. Almost optimal error rates are achieved approximately after 50 sessions or 15 thousand effective keystrokes and error rates double that figure are achieved after half the sessions or effective keystroke count.



BioType demo included with WebDNA 8.2


Here is how to implement BioType

DOWNLOAD WEBDNA NOW!

Top Articles:

F.A.Q

A compilation of some user's questions...

WebDNA Libraries

A list of available libraries for WebDNA...

Technical Change History

This Technical Change History provides a reverse chronological list of WebDNA changes...

[biotype]

BioType is a behavioral biometrics WebDNA function based on ADGS research and development (from version 8...

Tips and Tricks

A list of user-submitted tips ...

WebDNA Modules

A list of the currently available modules...

Related Readings:

[deletefolder]

Deletes a folder from your website...

[commitdatabase]

[COMMITDATABASE db=name...

[setmimeheader]

Causes WebDNA to add a new MIME header to the outgoing HTML...

[copyfolder]

Copy a folder and all its contenton your webspace...

[scope]

Explicitly define a block of WebDNA code that has a separate variable space...

[listdatabases]

Lists all the currently-open databases...