110
Internet of Things (IoT)
password as raw data in keystroke dynamics technique. Then some common features are
calculated by the following equations:
Key hold duration time (KD)=R i −P i
(6.1)
Interval time between two subsequent keys released (RR)=R i+1 −R i
(6.2)
Interval time between two subsequent keys pressed (PP)=P i+1 −P i
(6.3)
Interval time between one key released and next key pressed (RP)=P i+1 −R i
(6.4)
Interval time between one key pressed and next key released (PR)=R i+1 −P i
(6.5)
Interval time between first key pressed and last key released (t-time)=R n −P 1
(6.6)
Interval time between one key pressed and third key released
(Tri-graph-time)=R i+2 −P i
(6.7)
Interval time between one key pressed and fourth key released
(Four-graph-time)=R i+3 −P i
(6.8)
In our experiment, we have used only KD, DD, and UD for Dataset A, B, and C, respectively, as typing features, but we have used all the above features while working with
Dataset D which we have extracted using Equations 6.1 to 6.8.
Key pressure, finger tips size, finger movements, choice of control keys, type of frequent
errors, and choice of error correction mechanisms also can be measured for better performance in identification/authentication [5]. As per the study, these features could discriminate the gender and age group as well.
6.6.2 Normalization and Feature Subset Selection
Normalization is the first preprocessing step where we standardized the data within the
range {−1, 1} for faster processing. Feature selection method is used to find out the optimal
or close to optimal subsets of features when some irrelevant features are captured. It optimizes the accuracy rate along with computational speed. However, in our study, we have
not used any feature selection methods.
6.6.3 Gender and Age Group Recognition
We have evaluated 11 leading machine learning methods and calculated the accuracy listed
in Tables 6.3 and 6.4. We have divided total instances into 10 folds for cross-validation; here,
in each stage, 1 fold will be treated as a test set and others will be treated as training sets.
All the evaluation processes have been done with the supplied default parameter values
by Weka. Then, we have added this additional information to each sample as additional
features by assigning 0 for male and 1 for female; 0 for the age group 18–30 years and 1 for
the age group 30+ years to learn about the system manually.
SVMs a popular supervised machine learning method has been introduced by Vapnik
et al. [29] in 1995. Nowadays, SVMs have been widely studied in recognition and classification techniques to balanced datasets and have shown tremendous success in handwriting
Internet of Things (IoT)
password as raw data in keystroke dynamics technique. Then some common features are
calculated by the following equations:
Key hold duration time (KD)=R i −P i
(6.1)
Interval time between two subsequent keys released (RR)=R i+1 −R i
(6.2)
Interval time between two subsequent keys pressed (PP)=P i+1 −P i
(6.3)
Interval time between one key released and next key pressed (RP)=P i+1 −R i
(6.4)
Interval time between one key pressed and next key released (PR)=R i+1 −P i
(6.5)
Interval time between first key pressed and last key released (t-time)=R n −P 1
(6.6)
Interval time between one key pressed and third key released
(Tri-graph-time)=R i+2 −P i
(6.7)
Interval time between one key pressed and fourth key released
(Four-graph-time)=R i+3 −P i
(6.8)
In our experiment, we have used only KD, DD, and UD for Dataset A, B, and C, respectively, as typing features, but we have used all the above features while working with
Dataset D which we have extracted using Equations 6.1 to 6.8.
Key pressure, finger tips size, finger movements, choice of control keys, type of frequent
errors, and choice of error correction mechanisms also can be measured for better performance in identification/authentication [5]. As per the study, these features could discriminate the gender and age group as well.
6.6.2 Normalization and Feature Subset Selection
Normalization is the first preprocessing step where we standardized the data within the
range {−1, 1} for faster processing. Feature selection method is used to find out the optimal
or close to optimal subsets of features when some irrelevant features are captured. It optimizes the accuracy rate along with computational speed. However, in our study, we have
not used any feature selection methods.
6.6.3 Gender and Age Group Recognition
We have evaluated 11 leading machine learning methods and calculated the accuracy listed
in Tables 6.3 and 6.4. We have divided total instances into 10 folds for cross-validation; here,
in each stage, 1 fold will be treated as a test set and others will be treated as training sets.
All the evaluation processes have been done with the supplied default parameter values
by Weka. Then, we have added this additional information to each sample as additional
features by assigning 0 for male and 1 for female; 0 for the age group 18–30 years and 1 for
the age group 30+ years to learn about the system manually.
SVMs a popular supervised machine learning method has been introduced by Vapnik
et al. [29] in 1995. Nowadays, SVMs have been widely studied in recognition and classification techniques to balanced datasets and have shown tremendous success in handwriting