Document 7416387

Download Report

Transcript Document 7416387

Continuous Authentication Using
Keystroke Dynamics
by Bruno Guerreiro
BIOMETRICS
 Biometrics is the science of people
identification.
 This technology verifies the identity of an
individual based on his/her distinct
characteristics.
 There are two biometric authentication
methods:


24/04/2020
Physiological authentication
Behavioral authentication
Bruno Guerreiro
2
KEYSTROKE DYNAMICS
 Biometric behavioral method
 Checks the identity of an individual
 Monitored keyboard
 Typing characteristics are recorded in a form
of timings
24/04/2020
by Bruno Guerreiro
3
CHARACTERISTICS (TIMINGS)
 KEY DOWN | 5,69105626553095 | I
 KEY UP | 5,805169372085 | I
 KEY DOWN | 6,4678342689764 | N
 KEY UP | 8,5487345268945 | N
Duration 1
KEY1_DOWN
24/04/2020
Latency
KEY1_UP
Duration 2
KEY2_DOWN
by Bruno Guerreiro
KEY2_UP
t
4
AUTHENTICATION
 When a person gives us something that is
supposed to identify him/her and we check if
it is true or not.
 There are 2 main authentication techniques:


24/04/2020
Static – Just at once at the start and it is valid
till the end of the session
Continuous – Made continuously during the
session
by Bruno Guerreiro
5
RESEARCH QUESTIONS
 Is it possible to continuously authenticate a
user during normal typing with the same
results as static authentication ?
 Which are the advantages and limitations?
24/04/2020
by Bruno Guerreiro
6
EXPERIMENTS
 Experiment 1 (Enrollment phase) – Data
collection while typing a predefined text
 Experiment 2 (Analysis) – Data collection
while the user uses his computer free for
some days to compare to the previous
template afterwards
24/04/2020
by Bruno Guerreiro
7
EXPERIMENT 1
 Features

Single keys



E, I, N, O, S, T, L
Space, Backspace, Comma, Backspace
Double keys

MA, IN, IC, IS, BY, OF, TE, BE, NG, AN, AT, CO,
OR, LL, TT, Space+M
 Timings stored in one text file
24/04/2020
by Bruno Guerreiro
8
EXPERIMENT 1
24/04/2020
by Bruno Guerreiro
9
TEMPLATE (1)
 General
 Personalized
24/04/2020
by Bruno Guerreiro
10
EXPERIMENT 2
 Ran in
background
 Capture just some
features
 Less controlled
environment
 During a few days
 Timings stored in 2 text files:
 singleKeys.txt


E|-|4,646932|4,742566|dayX
doubleKeys.txt

24/04/2020
A|N|4,944549|5,0242685|5,0396254|5,1030760|dayX
by Bruno Guerreiro
11
DISTANCE METRIC
 Simple:
d Key k 
tk   k
k
 Gives the distance between two typing
patterns
 Normal or not
24/04/2020
by Bruno Guerreiro
12
DECISION RULE 1
 Single key:
if  d Key  k  TDur   D  0, else  D  1
 Double key:
if  d Key1 k  TDur   D1  0, else  D1  1
if  d Latency  TLat   D2  0, else  D2  1
if  d Key 2 k  TDur   D3  0, else  D3  1
if  D1  D2  D3  2  D  0, else  D  1
24/04/2020
by Bruno Guerreiro
13
ANALYSIS (1)
 Template VS experiment 2 data
0,061071127 … 0,089815773
24/04/2020
New templates needed !
by Bruno Guerreiro
14
ANALYSIS (3)
 With decision = 0/1
 7825 collected
Templates vs Person 5
Mistakes (%)
100
80
60
Person 5
40
20
16,4
0
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29
features
 1276 mistakes
 24 missing
features in the
template
 0 below
Templates
24/04/2020
by Bruno Guerreiro
15
ANALYSIS (4)
 With decision = 0/1
 10212 collected
Templates vs Person 10
Mistakes (%)
60
50
40
30
Person 10
16,8
20
16,8
10
16,2
15,9
0
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29
features
 1712 mistakes
 9 missing
features in the
template
 3 below
Templates
24/04/2020
by Bruno Guerreiro
16
DECISION RULE 2
 Single key:
if  d Key  k  TDur   D  0, else  D  d Key  k  TDur
 Double key:
if  d Key1 k  TDur   D1  0, else  D1  d Key1 k  TDur
if  d Latency  TLat   D2  0, else  D2  d Latency  TLat
if  d Key 2 k  TDur   D3  0, else  D3  d Key 2 k  TDur
D
24/04/2020
( D1  D2  D3 )
3
by Bruno Guerreiro
17
ANALYSIS (5)
 With new decision rule
 10212 collected
Templates vs Person 10
Mistakes (distance)
60
50
40
30
Person 10
20
10
9,9
0
1
3
5
7
9 11 13 15 17 19 21 23 25 27 29
features
 1712 mistakes
 9 missing
features in the
template
 0 below
Templates
24/04/2020
by Bruno Guerreiro
18
RESULTS
 With decision = 0/1:


FAR = 22 %
FRR = 0 %
 With decision = Distance – Threshold:


24/04/2020
FAR = 16 %
FRR = 0 %
by Bruno Guerreiro
19
COST FUNCTION
 Initial condition: C  0

When mistake:

When correct:

Every second:
24/04/2020
YES or NO
by Bruno Guerreiro
20
EXPERIMENTS SUMMARY
 Hard to find participants
 Some participants quitted
 Can I use Ctrl+C Ctrl+V ?
 Some problems with the Splitter line:

SPLITER|Username|09/05/2007 10:34:55
 Big gap between single and double key
number
24/04/2020
by Bruno Guerreiro
21
CONCLUSION
 Results shown a good potential.
 Duration timings shown to be more consistent
than latency timings.
 Highly sensitive to the used language to type.
 Typing characteristics depend on the
environment.
24/04/2020
by Bruno Guerreiro
22
FUTURE WORK
 More participants and longer period of time;
 Better experiments setup;
 Personalized choice of features;
 Improve distance metric / decision rule;
 Include other feature types, time of the day;
 Add key pressure;
 Include mouse movements;
24/04/2020
by Bruno Guerreiro
23