Automated login method selection in a multi modal authentication - login method selection based on user behavior

Automated Login Method Selection in a Multi-
modal Authentication System
Login Method Selection based on User Behavior
CHANG PEI SHAN
Security Information Lab
Mimos Berhad
Kuala Lumpur, Malaysia
ps.chang@mimos.my
DAHLIA DIN
Mimos Berhad
dahlia.din@mimos.my
WONG HON LOON
Mimos Berhad
hl.wong@mimos.my
SEA CHONG SEAK
Mimos Berhad
cssea@mimos.my
LEE KAY WIN
Mimos Berhad
kw.lee@mimos.my
Abstract² $Q LQWHOOLJHQW PRGHO LV SURSRVHG WR
DXWRPDWHVHOHFWLRQRIORJLQDXWKHQWLFDWLRQPHWKRGLQD
PXOWLPRGHO DXWKHQWLFDWLRQ VVWHP EDVHG RQ XVHU
EHKDYLRU SURILOH 7KH SURSRVHG SURWRFRO DQDO]HV WKH
XVHUEHKDYLRUGDWDWRPLQLPL]HWKHSURFHVVLQUHDOWLPH
DQGSUHYHQWXQWUXVWHGDWWHPSW PHDQZKLOHLWDEOHWR
IDFLOLWDWHIULFWLRQOHVVXVHUH[SHULHQFHLQDPXOWLPRGDO
DXWKHQWLFDWLRQ VVWHP 7KH SURSRVHG VVWHP
GHWHUPLQHV D XVHU DW D ORJLQ LQWHUIDFH DQG UHWULHYHV
WKHXVHU¶VEHKDYLRUDOKLVWRULFDOGDWDDVLWVGHWHUPLQDQW
IDFWRUV PDWFKLQJ XVHU EHKDYLRU EDVHG RQ GDWD
UHWULHYDOIURPGDWDEDVH:KHUHLQWKHEHKDYLRUDOGDWD
LQFOXGHV WKH XVHU¶V ORJLQ VHVVLRQV WLPH JHRORFDWLRQ
DFFHVVHGDSSOLFDWLRQDQGWSHRIXVHUDJHQWXVHG7KH
VVWHPSURFHVVHVWKHEHKDYLRUDOGDWDWRJHQHUDWHDXVHU
SURILOHEDSURILOLQJPRGXOHDWWKDWSRLQWHYDOXDWHV
WKH XVHU¶V SURILOH E IDFWRULQJ LQ HQYLURQPHQWDO
SDUDPHWHUV RI WKH XVHU WHUPLQDO WR VHOHFW DQ
DXWKHQWLFDWLRQPHWKRGEDQHYDOXDWLRQPRGXOH7KH
VHOHFWHG DXWKHQWLFDWLRQ PHWKRG ZLOO EH GLVSODHG IRU
WKHXVHUWRFRPSOHWHWKHDXWKHQWLFDWLRQSURFHVV2QWKH
RWKHU KDQG WKH VVWHP ZLOO XSGDWLQJ WKH XVHU SURILOH
ZLWKGDWDUHODWLQJWRDQHZVXFFHVVIXOORJLQVHVVLRQIRU
IXWXUHHYDOXDWLRQ
Keywords - Automated Method Selection, Multi-
modal Authentication system, User Historical Profile,
Profiling Module, Environmental Parameters,
Evaluation Module, Normalized Distribution
I. INTRODUCTION
In the past couple of years, we have witnessed the rapid
development of technology especially on cloud and
network computing and it has significant impact on the
world today. Users enjoy the technology to reduce the
operating time, improve operation efficiency and other
convenience by registered and stored all their confidential
personal data in cloud application or others web-based
programs. Unfortunately, the vast majority of network
attacks are initiated, the cloud terminal of the weight loss
caused by the end user security environment weakened, the
client is more susceptible to viruses and Trojans infected,
the user password is stolen incident has increased, all of
this lead to face with more severe information security
challenges [1].
Hence, the user authentication is essential requirement
for current protected web-based applications nowadays.
The common ways of authentication is through passwords
(knowledge-based security), fingerprint or face (inherence
factors) and ID cards (token based security) which are
normally accustomed limit access to a variety of systems
[2]. Initially, a traditional authentication mechanism
provides a sole authentication method for user to proceed
in a secure web page access. However, current
development in information technology manages to
decouple authentication function from application, thus
allowing for new possibilities for the application to counter
the internet fraud in a more cost effective way [3]. In this
case, the traditional identity authentication and content
authentication cannot meet the basic security needs of the
web-based application and cloud computing environment
[4].
Therefore, multifactor authentication is required which
is a security system that requires more than one method of
authentication from independent categories of credentials
to verify the user’s identity for a login or other transaction
[5]. Then again, to let user must aware and able to provide
“what the user has” and “what the user knows”, a multi-
modal authentication system called MIMOS Unified
Authentication Platform (Mi-UAP) is being introduced by
author’s team member. Multi-modal means, multiple
authentication methods are provided for user to access a
web application. Stand arise this multi-model
authentication system, user able to select a login option
they confident to provide for the current situation for
completing a successful authentication. Furthermore, the
current development phase is to generate an automated
selection of login method for user in a multi-modal
authentication system for facilitating frictionless and
improve user experience.
The rest of the paper is organized and structured as
follows. In Section II, we look into the background of our
authentication system Mi-UAP platform. We define the
related work by introduce the experimental environment
and user historical data analysis from our production in
Session III. For session IV, we introduce solution by
describe the architecture and flow of our design idea, as
120
978-1-5386-8546-4/19/$31.00 ©2019 IEEE

well as the analysis results based on collected historical
data. In Section V, we include the major findings relate to
the analysis we have done. Finally, we describe the
benefits and limitations of the proposed protocol, also the
future work in the conclusion section.
II. BACKGROUND
A. Platform Architecture
The organization Information System Security
research lab has developed a multi-modal authentication
platform named Mi-UAP to provide a convenient and
user-friendly service for both organizations and individual
users [6]. Mi-UAP provides an infrastructure which able
deliver authentication services to applications with the
authentication mechanism decoupled from application
implementation. This is a self-managed system which
incurs zero administrative effort [3]. Besides, a user has
an option to use any login method based on user’s
preference as six authentication login options is available
for users to select.
The authentication method implemented in our
platform depicted in Figure 1 includes four common
authentication methods used today which is password,
certificate, One Time Password (OTP) token and OTP
SMS; and other two proprietary authentication methods
was introduce and developed by the authors which is
Time-Constrained Key (TCK) and 2DBarcode. The
development and production spanned for eight years. The
authentication methods are progressively added, case by
case basis, either as new project requirements or to solve
problems raised from previous authentication method [4].
)LJ0L8$33ODWIRUP$UFKLWHFWXUH
The benefits of the authentication platform utilizes
Single Sign-On (SSO) for a seamless environment
operation and adaptive authentication with threat response
without modifying existing applications [6]. Mi-UAP also
able support two factors authentication, which following a
successful first authentication, a second single
authentication method is displayed to challenge user in the
new single web page. The goal of two factor
authentication is to create a layered defense and make it
more difficult for an unauthorized person to access a
target
B. Problem Statements
While providing flexibility to the users, multi-modal
authentication can lead to some adverse effects:
1. Confusion and annoyance as unregistered
authentication methods are shown in front of user.
2. Undesirable friction in user experience which need
to perform an additional step to choose the
authentication method as others common
authentication system doesn’t behave the same.
)LJ/RJLQ2SWLRQVLQ0L8$3
Refer to Figure 2, the multi-modal authentication
system shows a list of login methods when user tries to
login their application. There have been pass experiences
where user forget the methods they registered on the
particular gateway. Hence, user has to spend some times
for the login process to recall their registered login
method and credential.
Accordingly, it would be desirable and convenient to
provide a system that can automatically select a login
method for a user based on the behavior characteristics of
a current session when compared to the user’s historical
profile.
III. RELATED WORK
A. The Environment
For the authentication platform system, it is supported
by a physical host with 8 mandatory Virtual Machines
(VM) with different functionality as shown in Figure 1.
For the physical host, it is using a 24-core CPU, 96B
RAM and 1.1TB disk storage. All the Virtual Machines
(VM) are deployed Ubuntu version 16.04LTS as the
operating system.
B. Collected Data Based on User Behavior
This paper is an industry case study of the development
on multiple authentication methods as an authentication
service for 16638 registered users. This case study is
unique because the data is extracted from a live
production system with registered users inside multiple
system logs. Since this paper emphasizes on user behavior
for multi-modal authentication, related analytic is
measured based on scenario of users performing
authentication to access the production secure portal. The
authentication system log contained the information such
as process time, authentication method selected,
application which to access, geolocation and user agent.
The log duration is 3 months, which is from July 2018 to
Sept 2018. As review purpose in this paper, 1 user who
named as User C is selected for the user behavior
analytics based on 100 successful access from
authentication logs last 3 months.
121

7DEOH8VHU3UHIHUHQFH0HWKRGEDVHGRQ%URZVHU
User Agent Selected Authentication Method
Password TCK OTP Token Certificate
Firefox 12 5 8 28
Internet Explorer 5 0 0 0
Chrome 31 9 2 0
Based on Table 1, User C prefers to use Firefox as default
browser with Certificate login method. Secondly is
Chrome, while prefer login method is password. It might
because the user certificate is stored under Firefox.
7DEOH8VHU3UHIHUHQFH0HWKRGEDVHGRQ$SSOLFDWLRQ
Application Selected Authentication Method
Myprofile 27 0 0 3
MyEss 11 14 10 25
IDPAA 10 0 0 0
Table 2 shows that User C used to access MyEss
application most of the time with certificate login method.
Whereas, password is the prefer method of User C when
go to others application.
7DEOH8VHU3UHIHUHQFH0HWKRGEDVHGRQ/RFDWLRQ
Location Selected Authentication Method
Geolocation 1 44 10 10 27
Geolocaion 1: Kuala Lumpur Geolocation 2: Petaling
Jaya
Geolocation 3: Seri Kembangan
Table 3 shows that User C normally access from
Geolocation 1 by using both favorite login method which
is password and certificate.
7DEOH8VHU3UHIHUHQFH0HWKRGEDVHGRQ7LPHVWDPS
Timestamp Selected Authentication Method
Timestamp 1 23 7 4 19
Timestamp 2 17 4 6 8
Timestamp 3 3 3 0 6
Timestamp 4 0 0 0 0
Timestamp 1: 6am – 12pm Timestamp 2: 12pm – 6pm
Timestamp 3: 6pm – 12am Timestamp 4: 12am – 6am
Table 4 indicates that User C access application mostly on
Timestamp 1 with password login method follow by
certificate login method.
IV. SOLUTION
Figure 3 is a full diagram illustrating a representative
environment in which an automated method selection in a
multi-modal authentication system may be implemented.
)LJ3URSRVHG)UDPHZRUN'LDJUDPV
The paper’s idea discloses a computer-implemented
method to automatically select an authentication method
for a user based on a plurality of historical profiles
corresponding to user identifier in a multi-modal
authentication system. In another aspect, a protocol that
comparing behavioral characteristics of the user during
current session with the user behavioral profile that
previously developed based on prior usage patterns of the
user through the historical login session. User behavior
profile included device location, usual login timestamp,
internet service provider and application.
)LJ2YHUDOO3URFHVV)ORZ
Figure 4 is the overall process flow of the proposed
protocol in this paper. First, user comes to an
authentication gateway to login for an application. The
authentication system will ask the user to enter her/his
username first instead of show a list of login method to let
user select. After user enters their username, the system
only will start the enable the automated method selection
flow if the username is verified by the system. The
authentication system automatically selects a login method
for the user based on a plurality of historical profiles
related to his/her username inside the database. If the user
continues and completed login with the selected method,
the successful login attempt is added to the historical
profiles.
Alternatively, user may wish to skip the selected
method, at that moment, the automation authentication
system will evaluate the profile of user again by exclude
the skipped method as show in Figure 5.
122

)LJ3URFHVV)ORZZKHQXVHUVNLSSHGVHOHFWHGPHWKRG
In some scenarios, two or more methods may have the
same plurality of historical profiles; hence, the system will
pick the last used method from most recent login based on
the timestamp table which will explain more in the
following session.
Basically, the proposed process flow shows in Figure 3
depends on 2 modules as following:
1. 3URILOLQJ 0RGXOH, wherein processing the
behavioral data includes the user’s login sessions,
geolocation, and type of browser used to generate
a user profile
2. (YDOXDWLRQ0RGXOH, wherein evaluating the user’s
profile by factoring in environmental parameters
of the user terminal to select an authentication
method.
A. Profiling Module.
Profiling process extracts the data that is received,
stored and transmitted by authentication server after user is
verified. Successful login data are profiled based on the
environmental parameters, e.g. access application,
browser, and geolocation.
Table 5 shows the normalized distribution implemented
on a profile based on user agent from User C from Table 1.
Normalized distribution equation are obtained from below
based on password method in Firefox browser.
=
= 0.226
7DEOH1RUPDOL]HG'LVWULEXWLRQRI8VHUEDVHGRQEURZVHU
User Agent Selected Authentication Method
Firefox 0.226 0.094 0.152 0.528
Internet Explorer 1 0 0 0
Chrome 0.738 0.214 0.048 0
On the other hand, the profiling module also stores the
timestamp table of each method for every successful login
attempt as illustrated in the table 6 below. It is stored to use
when system obtain more than one favorite method for
user after the evaluation module.
7DEOH/DVW6XFFHVVIXO/RJLQ7LPHVWDPSRIHDFKPHWKRG
3DVVZRUG 7. 2737RNHQ HUWLILFDWH
30/9/2018 12/9/2018 31/8/2018 29/9/2018
1.39pm 4.50pm 9.15am 10.02am
B. Evaluation Module
Automated authentication system based on user
behavior profile then the system analysis an overall
distribution for the current session user. After analysis the
highest normalized distribution, one of login method
appear which is based on user-defined preference
regarding their respective apparatus activities. For example
A, if User C in Seri Kembangan and wish to access
application MyEss by using Chrome browser on 1pm, the
overall distribution of each method, is calculated as
example equation below and tabulated in Table 8.
Overall distribution on Password Method
= 0.738x0.183x0.484x0.486 = 0.0318
7DEOH2YHUDOO'LVWULEXWLRQRIHDFK0HWKRGLQ([DPSOH$
Environmental
Parameters
Authentication Methods
Chrome 0.738 0.214 0.048 0
MyEss 0.183 0.233 0.167 0.416
Geolocation 1 0.484 0.11 0.11 0.3
Timestamp 2 0.486 0.114 0.171 0.229
Overall
Distribution
0.0318 0.0006 0.0002 0
Based on table 7, Password method is selected for user
C to login as it has the highest normalized distribution
(HND) when user C having situation in Example A.
)LJ(YDOXDWLRQ3URFHVV)ORZZKHQVHOHFWHGPHWKRGLVVNLSSHG
Alternatively, for some scenario as shown in Figure 6,
if more than one method having same highest normalized
distribution (HND), the automated authentication system
will select the recent last use method based on the
timestamp table which stored during every successful login
attempt of the user in profiling module as shown in Table 6
and the overall distribution cannot equal to zero.
Otherwise, system will directly show the HND login
method only.
However, user is allowed to skip the selected method.
If user skips the selected method but the user only having
one login method, then system will show related error to
user. While if user has activated more than one login
method previously, the system will exclude the skipped
method and do evaluation again to select another login
method for user. Hence, if user do not skip the selected
method and continue login, the related environment data
will stored by the authentication server to keep as profiling
module.
As in Example A. if user C skipped the Password
Method, automated authentication system will select TCK
as next login method for User C as TCK method having
the latest login, while Password method is excluded and
overall distribution of certificate method is equal to zero.
Normalized Distribution of
Password login in Firefox
123

V. DISCUSSION
The aim of this paper is to attempt an intelligent
method which allows user the flexibility to select its own
preferable login method when access a restricted web
service [7].
From the finding in Table 1 to 4 shares that every
user have their own behavior when they access the
authentication system in different authentication factor
such as location, timing, application to access, user agent
and etc. These end-user behavior is credible data analytic
to mature one of the important research content for our
paper. The findings from Table 5 to 6 which involve the
normalized distribution formula indicate user C favorite
methods are password and certificate while depends on
which browser is using. For example, the system will
prompt user C certificate login method when user C
access service with Firefox browser. The results clearly
show that the automated authentication system able to
select user prefer login method in different scenarios
based on the two module which is profiling and evaluation
as shown in session IV.
VI. CONCLUSION
Clearly, the requirement for reliable techniques for user
authentication has enhanced within the wake of
heightened considerations regarding security and fast
advancements in communication, quality and networking
[8]. A large kind of applications need reliable verification
schemes to verify the identity of a person requesting their
service.
Current proposed protocol is to provide user login
method automatically based on user’s historical profile to
facilitate the multi-modal authentication process of a user.
At the same time, it able to proliferate the security level
which based on user historical behavior data. Firstly, the
authentication system automatically selects a login method
for the user based on a plurality of historical profiles
related to his/her username inside the database. If the user
continues and completed login with the selected method,
the successful login attempt is added to the historical
profiles.
As summary of the results we discussion in session V, it
indicates that the proposed approach is able to select login
method for user automatically based on their historical
behavior, it positively facilitates the undesirable friction in
user experience. The results also show that confusion and
annoyance will not occurred as unregistered authentication
methods are shown.
However, the automated authentication system do have
limitation especially for new user. It is because no
behavior data able to collect since the user is newly
registered. Hence, the automated authentication system
cannot proceed to evaluation module. Furthermore, the
current proposed protocol is based on the user historical
behavior from first day register, so it may not precise for
users that change their recent behavior which possible
caused by many environment factor such as change laptop,
work relocation and etc.
For future work, we are planning to deploy the proposed
protocol in this paper, along with its implementation and
experimentation in real testbeds. Correspondingly, the
team will study to design a better approach for increasing
the design complexity and overcome the limitations but at
the same time, remains improving the end-user experience.
Conclusively, when this application is fully deployed,
the intelligent protocol will be able to prompt user favor
login method based on their historical behavior. This will
save time for user while still able provide positive
identification, also, the multi-modal authentication system
will serve as a data repository for evaluating the user
historical behavior.
ACKNOWLEDGMENT
We acknowledge the support provided by Dr Cheong
Hoon Sin for providing the advice on the idea and
framework related of this paper. Also, Mr. Wong Hon
Loon to provide the authentication logs from the
production platform as reference on user behavior analytics
to accomplish this paper.
REFERENCES
[1] Chengyuan Zhang, Haishui Xu, “Research on user behavior
authentication model based on stochastic Petri nets”, Proceeding of
the American Institute of Physics
[2] Garg, Suneet vig Savita gupta, Renu, “Multimodal
Authentication System: An Overview”, International Journal of
Control Theory and Applications, Vol 10, Page 111 to 119, 2017
https://www.researchgate.net/publication/319292208_Multimodal_
Authentication_System_An_Overview
[3] Sea Chong Seak, Ng Kang Siong, Wong Hon Loon, Galoh
Rashidah Haron, “A Centralized Multimodal Unified
Authentication Platform for Web-based Application”, WCECS
2014, Proceedings of the World Congress on Engineering and
Computer Science
[4] Galoh Rashidah Haron, Dharmadharshni Maniam, Latifah Mat
Nen, Nor Izyani Daud, “User Behaviour and Interactions for
Multimodal Authentication”, PST2016, published by IEEE.
[5] Margaret Rouse, “Multifactor authentication (MFA)”, March 2015,
https://searchsecurity.techtarget.com/definition/multifactor-
authentication-MFA
[6] Sea ChongSeak, Chang PeiShan, Wong HonLoon, Dahlia Din,
“Research Institution Software Development Process Improvement
to Produce High Quality Research Software Assessment on
Technical Software Package Installation”, ISAI2018, published by
IOP
[7] Anusiuba Ifeanyi, Anigbogu S.O., Onyesolu Moses, Okonkwo,
“Multimodal Authentication Techniques For Staff Identification
And Tracking”, December 2014, proceeding of West African
Journal of Industrial Academic Research.
[8] Stacy Lyn Stubblefield, “System and method for utilizing
behavioral characteristics in authentication and fraud prevention”,
15 March 2013, US9275211 B2
[9] David M. Grigg, Peter John Bertanzetti, Michael E. Toth, Carrie
Anne Hanson, “User authentication based on historical user
behavior”, 7 Feb 2014, US9185101 B2
[10] Zhengyou Zhang, David W.Williams, Yuan Kong, Zicheng Liu,
David Kurlander, Mike Sinclair, “Multimodal authentication”, 29
June 2005, US8079079B
124

Automated login method selection in a multi modal authentication - login method selection based on user behavior

More Related Content

What's hot (16)

Similar to Automated login method selection in a multi modal authentication - login method selection based on user behavior (20)

More from Conference Papers (20)

Recently uploaded (20)

Automated login method selection in a multi modal authentication - login method selection based on user behavior