IEEE Consumer Electronics Magazine - January/February 2022 - 30

AI-Driven Security Solutions for the Internet of Everything
Figure 4. Three stages of OP-Grid.
Figure 5. Data flow of the user-verification
algorithm.
list of different features extracted from touch
events and sensors user's OP-Grid pattern.
The SVM classifier was trained based on the
aforementioned features collected from different
users; the collected data are served as training data
and labeled as valid users (1). Once the algorithm
is trained, it determines whether the legitimate
user or some other user tried to login. SVM implementation
had been used provided by LIBSVM,10
which supports the estimation of probability.
Figure 4 shows an overview of the proposed
OP-Grid. The system consists of two components.
The OP-Grid interface is installed on smartphones.
Once the user draws the pattern by using OP-Grid
technique[seeFigure4(a)],itisverified by the
password verifier[seeFigure4(b)] if thepassword
is correct; then, it will move to the authentication
algorithmof the user [see Figure 4(c)].
User authentication algorithm, shown in
Figure 5, has two phases: the enrollment phase
Table 3. Comparison of the Related Work.
References
Algorithm
Kumar et al.11
SVM
(training) and the verification phase (testing).
In the former mode, the algorithm is fed with
the true patterns labeled with their usernames,
i.e., patterns that are verified by the password
verifier. From these patterns, the system creates
a profile of a user's pattern behavior,
which is required in the verification mode. In
the verification mode, the profile will be
matched with the patterns from an unknown
user. The algorithm will decide whether the
drawn pattern has been performed by the legitimate
user or not. The two classes required to
learn the SVM are, first, the positive samples
consisting of patterns from a valid user and,
second, the negative samples consisting of patterns
from all other users. Table 3 shows the
comparison of OP-Grid with existing schemes
that use the SVM classifier.
Findings
Limitation
The proposed SVM-HMM
outperform the NB and cosine
similarity.
Frank et al.12
SVM
The SVM is more accurate than
the K-nearest neighbor.
Bo et al.13
SVM
The proposed approach
requires less power and is
authenticated easily and
accurately.
Ehatisham-ul Haq et al.14
OP-Grid
SVM
SVM
The SVM performs better in
different cell phone positions
than the DT and K-nearest
neighbor.
Two-factor authentication using
fifth factor and SVM classifier.
Authentication by brain signal across
the network is limited to the
proposed framework.
Theproposed authentication scheme
does not take into account attributes,
such as position, accelerometer data,
etc.
The approach proposed neglects
contextual characteristics that can
affect the validity of the model.
The proposed approach examines
only a number of sensors and does
not take contextual information into
consideration.
User can draw only pattern of 90.
30
IEEE Consumer Electronics Magazine
IEEE Consumer Electronics Magazine - January/February 2022

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - January/February 2022
