108
Internet of Things (IoT)
6.4.7 Circumvention
How easy it is fraud specific biometric characteristics. Keystroke dynamics is not easy to
mimic even if you observe the typing style of others many times.
6.5 Benchmark Soft Biometric Datasets on Keystroke Dynamics
In this section, we describe the datasets used in our experiment. Many variants of authenticated datasets on keystroke dynamics are available on the Internet, which can be downloaded
or accessed on request. In this chapter, we have used four datasets for different predefined
texts as well as different environments for the prediction of gender and age group. Table 6.2
represents the details of the publicly available authentic and recognized datasets. All the
datasets are given by a name for the purpose of identification throughout this chapter.
We summarized the different keystroke dynamics datasets in Table 6.2 which are used
in our study. Most of the researchers have performed different experiments to develop
a model to recognize the user with these datasets. Some of them obtained impressive
results to identify the user, but they are not acceptable in practice. Datasets A, B, and C
were collected through a computer keyboard. Dataset A was collected from 38 male and
27 female users where 38 were aged 18–30 years and 27 were aged 30+ years; Dataset B
was collected from 25 male and 13 female users where 23 were aged 18–30 years and
15 users were aged 30+ years;. Dataset C was collected from 21 male and 21 female users
where 24 were aged 18–30 years and 18 were aged 30+ years; and Dataset D was collected
from 26 male and 25 female users using a touch screen device where 11 were aged 7–18
years, 30 were aged 19–29 years, and 10 were aged 30–65 years.
6.6 Proposed Methodology
Biometric systems are not 100% accurate as per Jain et al. [21] due to various problems in
data acquisition methods or interclass variations. As per previous studies, accuracy can
be improved by using the soft biometric information as additional features with the typing pattern in keystroke dynamics. Giot et al. [7] used gender as additional information to
predict gender with 91% accuracy. In order to realize this technique in practice, we have
used FRNN-VQRS to predict the gender as well as the age group of the users based on
the typing pattern of different predefined texts which elicited more than 94% accuracy on
CMU keystroke dynamics dataset. The performance metric, area under curve (AUC), is
always high and proved FRNN-VQRS to be an efficient approach. In this study, we applied
gender and age group as additional features and obtained 94.37% accuracy in keystroke
dynamics authentication which is a gain of 6.29% using the same algorithm as a recognition method for predefined text “.tie5Roanl.” The proposed methods are described in the
following subsections:
6.6.1 Data Acquisition and Feature Extraction
It is the most fundamental and essential part in any biometric system. Here, key press time
(P) and key release time (R) in millisecond unit are recorded while typing user ID and
Internet of Things (IoT)
6.4.7 Circumvention
How easy it is fraud specific biometric characteristics. Keystroke dynamics is not easy to
mimic even if you observe the typing style of others many times.
6.5 Benchmark Soft Biometric Datasets on Keystroke Dynamics
In this section, we describe the datasets used in our experiment. Many variants of authenticated datasets on keystroke dynamics are available on the Internet, which can be downloaded
or accessed on request. In this chapter, we have used four datasets for different predefined
texts as well as different environments for the prediction of gender and age group. Table 6.2
represents the details of the publicly available authentic and recognized datasets. All the
datasets are given by a name for the purpose of identification throughout this chapter.
We summarized the different keystroke dynamics datasets in Table 6.2 which are used
in our study. Most of the researchers have performed different experiments to develop
a model to recognize the user with these datasets. Some of them obtained impressive
results to identify the user, but they are not acceptable in practice. Datasets A, B, and C
were collected through a computer keyboard. Dataset A was collected from 38 male and
27 female users where 38 were aged 18–30 years and 27 were aged 30+ years; Dataset B
was collected from 25 male and 13 female users where 23 were aged 18–30 years and
15 users were aged 30+ years;. Dataset C was collected from 21 male and 21 female users
where 24 were aged 18–30 years and 18 were aged 30+ years; and Dataset D was collected
from 26 male and 25 female users using a touch screen device where 11 were aged 7–18
years, 30 were aged 19–29 years, and 10 were aged 30–65 years.
6.6 Proposed Methodology
Biometric systems are not 100% accurate as per Jain et al. [21] due to various problems in
data acquisition methods or interclass variations. As per previous studies, accuracy can
be improved by using the soft biometric information as additional features with the typing pattern in keystroke dynamics. Giot et al. [7] used gender as additional information to
predict gender with 91% accuracy. In order to realize this technique in practice, we have
used FRNN-VQRS to predict the gender as well as the age group of the users based on
the typing pattern of different predefined texts which elicited more than 94% accuracy on
CMU keystroke dynamics dataset. The performance metric, area under curve (AUC), is
always high and proved FRNN-VQRS to be an efficient approach. In this study, we applied
gender and age group as additional features and obtained 94.37% accuracy in keystroke
dynamics authentication which is a gain of 6.29% using the same algorithm as a recognition method for predefined text “.tie5Roanl.” The proposed methods are described in the
following subsections:
6.6.1 Data Acquisition and Feature Extraction
It is the most fundamental and essential part in any biometric system. Here, key press time
(P) and key release time (R) in millisecond unit are recorded while typing user ID and