See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/329499572

EEG-based Age and Gender Prediction Using Deep BLSTM-LSTM Network Model

Article in IEEE Sensors Journal · December 2018

DOI: 10.1109/JSEN.2018.2885582

CITATIONS READS
63 1,187

4 authors, including:

Anmol Gupta Partha Roy

Indian Institute of Technology Roorkee Indian Institute of Technology Roorkee
8 PUBLICATIONS 95 CITATIONS 244 PUBLICATIONS 4,648 CITATIONS

SEE PROFILE SEE PROFILE

Debi Prosad Dogra

Indian Institute of Technology Bhubaneswar
124 PUBLICATIONS 2,675 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Offline Handwritten Signature Identification and Verification View project

Video Frames Prediction View project

All content following this page was uploaded by Anmol Gupta on 07 October 2021.

The user has requested enhancement of the downloaded file.

 1

EEG-based Age and Gender Prediction

Using Deep BLSTM-LSTM Network Model
*Pallavi Kaushik, *Anmol Gupta, Partha Pratim Roy, Member, IEEE and Debi Prosad Dogra, Member, IEEE

Abstract—With the rapid development of brain computer electrical signals of brain using electrodes attached to the
interfaces (BCI), the number of applications that use BCI scalp. It is available in both invasive and non invasive forms,
technology is increasing thick and fast. Prediction of age and though the non-invasive devices are more popular due to
gender of a person through EEG analysis is a new application of
BCI that has been proposed in this paper. An industry standard affordable pricing and ease of use.
EEG recording device has been used to record cerebral activities Ageing refers to the process of inevitably growing older that
of 60 subjects (both male and female) in relaxed position with is genetically determined and evolutionary modulated. While
closed eyes. Deep BLSTM-LSTM network has been used to growing older, there are several changes to the Central Nervous
construct a hybrid learning framework for the aforementioned
analysis. Accuracy of 93.7% and 97.5% have been recorded System (CNS) which may lead to diseases like Alzheimer,
for age and gender classification problems respectively. These Dementia, Parkinson etc. The changes in the Central Nervous
values are better than the state-of-the-art methods. Our analysis System (CNS) can be diagnosed with the help of EEG signal
also reveals that beta band frequencies are better in predicting analysis [32]–[34]. As far as gender is concerned, male and
the age and gender as compared to other frequency bands of female brains differ functionally, anatomically [35]–[37], and
the EEG signals. The proposed method has several applications
including biometric, health-care, entertainment and targeted in the chance of development of the neuropsychiatric diseases
advertisements. and how they react to the treatment [38].
Index Terms—Age Detection, BCI, BLSTM, Deep Learning,
There are some studies which have used traditional machine
EEG, Gender Detection learning approaches like LDA, SVM, Random Forest etc. to
analyse the brain signals measured using EEG for various BCI
applications [1]–[4], [41]. The problem with the traditional
I. I NTRODUCTION approaches is that, some assumptions are made in order to
Brain Computer Interface or BCI is an emerging field extract the features that are relevant to that particular appli-
which allows humans to communicate with computing devices cation. These assumptions are based on the theories that may
without using conventional input systems. This communication not be true across all applications. As a result, some important
in turn allows several applications which assist humans to features may get excluded during traditional feature extraction
perform varying tasks.The applications can be broadly cat- process. However, deep learning model tries to learn features
egorized into medical and non-medical types. For example, that may not be possible to extract using traditional feature
a person with hearing impairment, vision issues or limited extraction techniques. In this paper, we propose a new method
physical movements can perform various tasks with the help to predict the age and gender of a person through EEG signals
of neuroprosthetics. The non-medical applications include recorded using a portable 14-channel EEG recording setup. Bi-
authentication [6], [7], controlling appliances, playing games directional LSTM has been used to predict the age and gender
[8], typing [9], etc. All such applications can be assisted of the person, which is possibly the first of approach of this
using brain signals. Human brain works with the help of kind and the framework is called as ”Deep BLSTM-LSTM”.
neurons that are connected with other neurons using axons and The remainder of this paper is organized as follows. Section
dendrites. The activity in our brain is a by-product of these II reviews the related work. Section III gives the methodology
neurons firing when a certain threshold of electrical energy is used for the dataset collection, pre-processing and architecture
achieved. Some of the electrical signals when travelling inside of the proposed Deep BLSTM-LSTM model. Section IV
the brain escape the protective layer of fatty white substance discusses the results for different models and compare them.
called the myelin, and can be recorded. These signals can Finally, we make a conclusion in Section V .
be interpreted for understanding. Cerebral activities can be The main contributions of the paper include,
recorded using EEG, ECoG, or fMRI devices. Out of these
1) Beta band contains more important information than
three EEG is less expensive and complex as per the recording
other bands for prediction of age and gender.
mechanism. Therefore, it is extensively used for developing
2) Construction of a deep hybrid framework using Bi-
BCI applications. EEG recording devices can measures the
directional LSTM and LSTM to map human cerebral
Pallavi Kaushik, Anmol Gupta and Partha Pratim Roy are with the Depart- activities with age and gender features.
ment of Computer Science and Engineering, Indian Institute of Technology, 3) Prediction of age and gender of human participants with
Roorkee, India. the help of above deep learning framework.
Debi Prosad Dogra is with the School of Electrical Sciences, IIT
Bhubaneswar, India. 4) Comparison of different LSTM, Bi-directional LSTM
*Both the authors have equal contribution. architectures and GRU.
 2

II. L ITERATURE R EVIEW and gender prediction. The study shows that appropriate model
initialisation via pretraining counteracts overfitting, leads to a
This section summarises various studies which are presently holistic perception of the input that essentially leads to better
available to interpret the relation between EEG wave signals performance. Majority of the existing research works compare
and age/gender. The difference in frequency of brain waves the results on the latest benchmark datasets such as Adience
as a consequence of age was first studied as early as 1939. [19] and lfw dataset 1 . The state-of-the-art accuracy in gender
Lindsey et al. [5] found that, there is an increase in the recognition is 92.8% and 62.8% for age detection.
emission of the waves with higher frequencies and reduction In [20]–[24] the authors have applied long short term mem-
in the waves with lower frequencies in various regions of ory (LSTM) [25] on the EEG signals in applications such as
the brain. Since then, several research works [28]–[31] have emotion detection, lapse in attention, seizure detection, brain
studied the effects of age and gender at different frequency activity etc. In [22], the authors present a study on the brain
bands of the EEG signals. The results of these studies suggest fog, i.e., the state when the brain is confused that results in low
that, in terms of age, delta and theta activity are possibly performance while doing tasks which require clear thinking.
the prominent bands in terms of brain activity until the age Detecting a confused state can lead to several applications
of four, while alpha and beta activities increase throughout in online education, offline education, driver fatigue detection
the childhood. Therefore, change in cerebral activity is highly etc. They have applied Bidirectional LSTM recurrent neural
dependent on age and hence there is a need to study such network on the EEG data to classify students’ confusion while
effects for various applications. Some of these studies [35]– watching online videos. They achieve the accuracy of 73.3%.
[38] also suggest that gender is responsible for the maturity In [45] Gupta et. al. detected emotions using EEG to
of brain signals. It has also been reported that girls maturity enhance the text users write on social media with 74.9%
typically lag in the EEG as compared to boys [44]. accuracy with data of 25 participants. Kumar et. al. [49] have
However, the use of machine learning for studying the used EEG in sentiment analysis. Khurana et. al. [50] have used
relationship between various statistical parameters and EEG the EEG signals for predicting how much a user is familiar
signals as a function of age, has been first studied by Paiva with certain English words. Yadava et. al. [51] have used the
et al. [4]. This study establishes the relationship between the EEG signals to predict whether a user likes or dislikes certain
EEG signals and age. EEG signals of 59 subjects was recorded products that are sold on an e-commerce website.
which were divided into 7 groups, ageing between 20 to In [26], Putten et al. have applied CNN on EEG data to
86 years from both gender. They used Linear Discriminant predict the gender of a person. It shows that fast beta activity
Analysis(LDA) for measuring the correlation between different (2025Hz) is the main distinctive feature for classifying gender
age groups and found that indeed there were significant and achieves accuracy as high as 80%. The authors have
differences among the age groups. collected data of 1308 subjects using 26-channel EEG with
In [1], Nguyen et al. have proposed a framework which uses 47% male subjects and 53% female subjects. As far as we
PARAFAC [2] and SVM [3]. They have used EEG features know, this is the only paper that has used deep learning on
in addition to speech-based features as inputs to the parallel EEG data to predict gender.
factor analysis. The features are, namely spectral power, Our paper emphasises on applying LSTM architecture in
relative power, and discrete cosine transform. They further use bidirectional modes to predict the age and gender which has
multilinear partial least squares (N-PLS) and SVM to improve not been done earlier.
the classification accuracy and recorded 92.5% and 93.8%
for age and two class gender classification. In [41], Kaur et III. M ETHODOLOGY
al. have used random forest for predicting age and gender.
EEG signals of 60 healthy subjects have been recorded using This section describes the dataset, preprocessing and the
14 channel Emotiv Epoch Plus and classification accuracy of proposed architecture of the Deep BLSTM-LSTM model
88.33% have been recorded for age and an accuracy of 96.66% which comprises of the BLSTM, LSTM and dense layers, for
has been achieved for gender classification. classifying the gender and age of an individual by analysing
In [10], Franke et al. have proposed a framework for pre- EEG signals. A comparison of different LSTM architectures,
dicting age of 650 healthy subjects aged between 19-86 years, namely single LSTM, Deep LSTM and BLSTM has also been
from MRI brain scans and used relevance vector machine provided in all five bands. The results have been compared
(RVM) [11] regression for the prediction which yielded a mean with raw EEG data. The overall framework is presented in
absolute error of 5 years. Fig. 1.
In [12]–[15], authors have used face images to predict
age and gender using deep learning, specifically different A. Description of the Dataset
versions of convolution neural network like GoogleNet [16], This work uses the dataset proposed in [41]. The dataset
CaffeNet [17], VGG-16 [18] or hybrid models. In [13], the comprises of the EEG signals of 60 individuals with 35 males
authors provide a comparative study on deep convolution and 25 females ranging from 6 to 55 years. The individuals
neural networks with their proposed method of initialisation, are divided into 6 age groups. A more detailed overview of
refereed to as Layerwise Relevance Propagation (LRP). The the dataset can be found in the Table II. The EEG signals
algorithm tries to understand the functioning of deep learning
by investigating which features are actually responsible for age 1 http://vis-www.cs.umass.edu/lfw/
 3

TABLE I: A summary of the existing research work in this domain.

Authors Type of Data & Technology Results

Lindsey et al. [5] EEG data, First attempt to establish the relationship between Activity decrease in the lower frequency regions while in-
EEG signals and age. crease in the higher frequency regions with age

Pavia et al. [4] EEG signals of 59 subjects aged between 20 and 86. Successfully classify the subjects based on the age group.
LDA was used for establishing the correlation among
groups.(Among the first to use Machine learning for the task)

Nguyen et al. [1] EEG data, used speech related features like spectral power Achieved 92.5% and 93.8% accuracy for age and two class
in addition to EEG features, Used SVM and N PLS classi- gender classification.
fication

Franke et al [10] MRI brain scans of 650 subjects, used RVM [11] for Regres- mean absolute error of 5 years.
sion

Kaur et al. [41] EEG signals of 60 healthy subjects, used Random Forest for Achieved 88.33% accuracy for age and 96.66% for gender
classification classification.

Lapuschkin et al. [13] Face images to predict age and gender using deep learning 92.8% accuracy in gender and 62.8% for age classification
(different Convolution Neural Networks)

Putten et al [26] The authors applied Convolution Neural Network on EEG 80% accuracy for gender classification.
data collected from 1308 subjects.

Fig. 1: A High-level representation of the functional block diagram of the proposed age and gender recognition system.

TABLE II: Dataset description

Age Group Number of individuals Males Females Class
6-10 10 5 5 0
12-15 10 6 4 1
18-23 10 6 4 2
25-29 10 6 4 3
33-38 10 6 4 4
42-55 10 6 4 5

genders and various age groups, are presented in [41].

Fig. 2: Position of the electrodes according to the 10-20
international system.
B. Preprocessing using Discrete Wavelet Transform (DWT)
were collected for 10 seconds and individuals were asked The EEG signals comprises of waves of various frequencies.
to sit in a relaxed position with their eyes closed. Emotive Neuroscientists have broken them into frequency bands with
Epoch plus device was used to capture the EEG signals using each frequency band responsible for a particular activity of
14-electrodes namely, AF3, AF4, F3, F4, F7, F8, FC5,FC6, the brain. Various frequency bands are also referred to as the
P7, P8, T7, T8, O1, O2 and 2 references (CMS and DRL) brain waves. The various brainwaves along with the activity
as per the International 10-20 system, shown in Fig. 2. The responsible for them are as follows:
brain maps showing the difference in the EEG signals for both • Delta- These are the waves that lie in the frequency range
 4

of 0.5-4 Hz and correspond to deep meditation and deep The Db-8 decomposition gives remaining four wavelet coef-
and dreamless sleep. ficients that correspond to noise and five wavelet coefficients
• Theta- These waves fall into the frequency range of 4-8 corresponding to alpha, beta, delta, gamma and theta brain
Hz and correspond to day dreaming, drowsiness or sleep. waves. The raw EEG data from an individual is shown in
• Alpha- These brainwaves lie between 8-12 Hz and cor- Fig. 3 along with corresponding DB-8 wavelet decomposition
respond to the resting state of the brain i.e calmness and outputs.
ideal relaxation.
• Beta- These waves lie within 12-30 Hz and corresponds
C. Adopted Deep Learning Architecture.
to active thinking, consistent focus, memory recall, high
alert and anxiety. Experiments have been conducted using varying setups of
• Gamma- These brain waves are comprised of the fre- LSTM and Deep BLSTM+LSTM models. We present two
quency > 30 Hz and they correspond to information pro- specific architecture of LSTM and BLSTM that have worked
cessing, cognition, learning and matching of recognized better than the other models in the present context.
objects, shapes, sounds, etc. 1) Deep Long Short Term Memory (LSTM): LSTMs [25]
As various activities of the brain correspond to a brain wave, are a form of Recurrent Neural Network (RNN) which over-
it makes sense to work with the individual brain waves rather come the issue of exploding and vanishing gradient in the
than the whole EEG signal depending on the type of activity RNNs. The building blocks of LSTM consists of a cell, an
one is trying to analyze. The Discrete Wavelet Transformation input gate, an output gate and a forget gate. The cell is
breaks the input signals into various constituting small range responsible for handling the long term dependency while the
frequency bands via decomposition of the input wave at three gates regulates the flow of values between the different
various levels and it also helps in analyzing the nonstationarity layers of the LSTM network.
in the signals. Experiments were done and the LSTM model that gave the
The DWT is done by obtaining the Approximation (A) best results for age and gender prediction is being described
and Detail (D) coefficients by performing decomposition at in this Subsection. The model consists of a LSTM layer
various levels. It is done through the use of low and high pass comprising of 128 hidden neurons. Next, a batch normalization
filters at various levels. The low pass filter (L) ignores the layer is applied which is further followed by another LSTM
high frequency fluctuations and helps in preserving the low layer with 64 hidden neurons and another batch normalization
frequency trends. Similarly, the high pass filter (H) helps in layer. A dense layer of 32 neurons follows these layers and
keeping the high frequency fluctuations and ignoring the slow finally it is connected to another dense layer comprising of 6
trends in the signal. The outcomes from the low pass filters neurons (age prediction) and 1 neuron for gender prediction.
help in forming the approximation coefficients and those from 2) BLSTM: Various BLSTM models were experimented
the high pass filters help in forming the detailed coefficients. with but this Subsection only describes the model that gave the
The Daubechies (Db)-8 wavelet transform has been used in best results. The model consists of a BLSTM layer followed
this work, hence the levels of decomposition are 8. by a dropout of 0.2 followed by batch normalization. A
The wavelet function at time t is shown in (1) and (2), dense layer of 32 neurons follows that is followed by another
where m = 0, 1, ..., M − 1, t = 0, 1, ..., T − 1, M = log2 (t), dense layer having 6 neurons for different age groups for age
n = 0, 1, ..., 2m − 1, T is the length of the signal, a0 and b0 classification and a single neuron for gender classification as
values are set to 2 and 1, respectively. it is a binary classification.
3) Proposed Deep BLSTM-LSTM: The Deep BLSTM-
Z ∞
LSTM architecture consists of three stacked LSTM layers
ψ(t) dt = 0 (1) with 256, 128 and 64 units, respectively. The first layer is
−∞
subjected to a dropout of 0.2 followed by batch normalization,
−m/2 while the rest of the two LSTM layers are accompanied with
ψm,n (t) = a0 ψ(am
0 t − nb0 ) (2)
batch normalization only. A dense layer of 32 neurons follows
The scaling and wavelet functions required for the evaluation that is followed by another dense layer having 6 neurons for
A and D coefficients given in (3) and (4), respectively. different age groups in case of age classification. It has a
single neuron in case of gender classification as it has only
φp,q (t) = 2p/2 h(2p t − q) (3) two classes, male and female. The model has been trained
for 50 epochs with 10-fold cross validation. The number of
ωp,q (t) = 2p/2 g(2p t − q) (4) trainable parameters has been found to be 528, 230 for age
and 527, 041 for gender prediction. The results obtained with
The Ai and Di coefficients at the ith level are evaluated using
the architecture are discussed in the next section.
(5) and (6), respectively.
After obtaining various brainwaves from the DWT trans-
1 X form, each brainwave is analyzed in order to know which brain
Ai = √ x(t)φp,q (t) (5)
T t wave is more responsible for determining the age and gender
of an individual.
1 X The EEG signals are first fed into a BLSTM layer of 256
Di = √ x(t)ωp,q (t) (6)
T t neurons and then into a dropout layer with dropout rate of 0.2.
 5

Fig. 3: Diagram showing the raw data and various brainwaves, namely delta, beta, theta gamma and alpha obtained after
DWT for a sample user taken from the dataset.

Fig. 4: Architecture of the proposed Deep BLSTM-LSTM model, with one BLSTM and two LSTM layers followed by
dropout, batch normalization and dense layers.

Next, batch normalization is performed and the results are then

fed into a LSTM layer comprising of 128 hidden neurons.
Another batch normalization layer is applied then which is
further followed by another LSTM layer with 64 hidden
neurons and another batch normalization layer. A dense layer
of 32 neurons follows these layers and finally it is connected
to another dense layer comprising of 6 neurons. This is used
for classifying the EEG signals for each age group. For
gender classification the same architecture is used with the
only difference of 1 neuron in the last dense layer. The total
number of parameters in the described model are 939, 750 with
Fig. 5: Results obtained using the brainwaves and the deep
938, 342 being trainable. Through experiments, it has been
LSTM model.
found that the beta waves give better classification accuracy
for determining both the age and gender of individuals. A
complete overview of the architecture is presented in Fig. 4. been used to experiment with each of alpha, beta, theta, delta
The model has been trained for 50 epochs. and gamma brain waves. Beta brainwave gives the highest
accuracy of 97.52% for gender as well as maximum accuracy
IV. R ESULTS of 93.69% for predicting the age of individuals using the
The training, testing and validation sets are composed of stacked Deep BLSTM-LSTM.
60%, 30% and 10% of the whole data. The Deep BLSTM- Experiments have been conducted on the raw dataset in
LSTM, BLSTM and LSTM models described above have order to draw the parallel between with DWT and without
 6

Fig. 6: Results obtained using the brainwaves via the

BLSTM model.

Fig. 8: Confusion matrix for age classification on the beta

brainwave using the Deep BLSTM-LSTM model.

Fig. 7: Results obtained using the brainwaves via the Deep

BLSTM-LSTM model.
DWT. Categorical cross-entropy and binary cross-entropy have
been used as the loss functions for age and gender prediction,
respectively. The accuracies of various brainwaves obtained
by using the deep LSTM, BLSTM and Deep BLSTM-LSTM Fig. 9: Confusion matrix for gender classification on the beta
models are as shown in the Fig. 5, Fig. 6 and Fig. 7. It can brainwave using the Deep BLSTM-LSTM model.
be seen that the alpha and beta waves produce better results
than the rest of the brainwaves and also that the prediction
results obtained from the raw data can be highly improved fication, respectively. They have smoothened the EEG signals
using the Discrete Wavelet Transform from 83.27% and 84.6% using the Savitzky-Golay filter of degree 2 and extracted
to 93.69% and 97.52% for age and gender classification, various features like mean, root mean square and energy
respectively. from the brainwaves obtained after the application of DWT.
The confusion matrix for the age prediction using the alpha Levi and Hassner [44] have proposed a Convolutional Neural
brain wave for the Deep BLSTM-LSTM model is shown Network for the prediction of age and gender of individuals
in Fig. 8. It can be seen from the Fig. 8 that the age by using a large dataset comprising of 26, 000 facial images.
group of 18 − 23 corresponding to class 2 has the least They reported the accuracy of 50.7% and 86.6% for age and
misclassifications, where only 2 individuals out of 3642 are gender classification, respectively.
categorised into the age group of 25 − 29 (Class 3). The We have also computed the classification accuracies using
maximum misclassifications can be seen in the age group of GRU (Gated Recurrent Unit) in order to establish the effec-
25−29, where 89 out of 2905 individuals have been classified tiveness of LSTMs to GRU. The architecture used is analogous
into the age group of 12 − 15 (Class 1). The confusion matrix to the Deep BLSTM-LSTM model, which gave the best
for categorising the gender using the beta brainwave is shown classification accuracies amongst the LSTM based models.
in Fig. 9, where the model misclassifies 38 male members Using GRU, beta brainwave gave an accuracy of 92.14% and
as females of the total 1469 males and it misclassifies 119 95.14% for age and gender classification respectively.
females of the total 3612 female individuals. We have also experimented with Convolutional Neural
The Deep BLSTM-LSTM model also gives better accuracy Networks (CNNs) and found the accuracies less than 30% and
than the state-of-the-art models for predicting the age and less than 50% for age and gender classification respectively.
gender of various individuals. Kaur et al. [41] have used the As CNNs are good at learning the features on thier own, we
random forest classifier on the same dataset and reported the performed the experiments on the raw EEG signals. CNNs
accuracies of 88.33% and 96.66% for age and gender classi- tend to work best on images and RNN models for temporal
 7

TABLE III: Comparative analysis with the recent state-of-the-art techniques.

Authors Methodology Dataset Accuracy

Kaur et al. [41] Random Forest for classifica- EEG signals of 60 healthy subjects, Age - 88.33% Gender- 96.66%
tion

Nguyen et al. [1] SVM and N PLS classification EEG data of 40 subjects in resting state Age-92.5% Gender-93.8%

Franke et al [10] RVM [11] for Regression MRI brain scans of 650 subjects, Correlation, r=0.92 between the predicted and
real age with the mean absolute error of 5
years.

Lapuschkin et al. [13] Convolutional Neural Face images Age-62.8% Gender-92.8%

Networks

Putten et al [26] Convolution Neural Network EEG data collected from 1308 subjects. Gender-80%

data, hence the LSTM, BLSTM, GRU and Deep LSTM- the vanilla neural network or traditional machine learning ap-
BLSTM perform much better than the CNN model for the proaches. Thus, we have developed a architecture called Deep
raw EEG signals. BLSTM-LSTM which uses one layer of BLSTM followed
by two layers of LSTM which in turn are followed by two
fully connected dense layers. The recorded accuracies are as
high as 93.69% and 97.5% for age and gender classification,
respectively. Moreover, no smoothing has been performed that
reduces computational burden and other overhead. However,
raw data is segregated into different frequency bands to analyse
how brain activity is affected with the change in age of a
person and whether there is a difference in the brain activity
among genders. It is found that, the results are in accordance
with the previous results mentioned in the literature. In our
results, beta band waves provide a better prediction accuracy
for gender and age classification.
In future, more data can be collected for the purpose of the
Fig. 10: Accuracies obtained for age classification using experiment. The age groups created can be expanded to 5-80
label encoding. years so that the elderly people can also use the application.
We have also tried to predict the age of the individuals Moreover, the technology can be mixed with several existing
by using label encoding. The same Deep BLSTM-LSTM and technologies to create new applications with the help of EEG
LSTM architectures described in the previous section have devices. For example, we can use the brain segmentation
been used to experiment. However, only the alpha and beta technology [46] to further improve the results and we can also
waves have been found to be giving better predictions for leverage the cloud computing facilities [52] for the purpose of
the age and gender prediction. It has been found that the sentiment analysis [47] and for better visualisation [48].
alpha brainwave gives the maximum accuracy of 91.31%. The
results obtained for beta and alpha brain waves on both the R EFERENCES
LSTM and Deep BLSTM-LSTM models are shown in Fig.
10. [1] Nguyen P, Tran D, Vo T, Huang X, Ma W, Phung D (2013). EEG- Based
Age And Gender Recognition Using Tensor Decomposition And Speech
Features., International Conference On Neural Information Processing,
632 - 639
V. C ONCLUSION AND F UTURE W ORK [2] Harshman, R. A., & Lundy, M. E. (1994). PARAFAC: Parallel Factor
Analysis. Computational Statistics And Data Analysis, 18(1), 39-72.
Predicting the age and gender of an individual can be highly
[3] Hearst, M. A., Dumais, S. T., Osuna, E., Platt, J., & Scholkopf, B.
useful for various applications such as health care, knowing (1998). Support Vector Machines. IEEE Intelligent Systems And Their
consumer preferences for online retailers, biometric, etc. As Applications, 13(4), 18-28.
EEG signals are unique for each person they carry a lot of [4] Paiva, L. R. M. D., Pereira, A. A., Almeida, M. F. S. D., Cavalheiro,
G. L., Milagre, S. T., & Andrade, A. D. O. (2012). Analysis Of The
meaningful information. We attempt to build a robust system Relationship Between EEG Signal And Aging Through Linear Discrimi-
for predicting the age and gender of an individual by analysing nant Analysis (LDA). Revista Brasileira De Engenharia Biomdica, 28(2),
the EEG signals. An EEG dataset of 60 individuals including 155-168.
[5] Lindsley, D. B. (1939). A Longitudinal Study Of The Occipital Alpha
male and female was recorded while they sat with their eyes Rhythm In Normal Children: Frequency And Amplitude Standards. The
closed in a resting position. As EEG signals are time series Pedagogical Seminary And Journal Of Genetic Psychology, 55(1), 197-
data where the value at time t + 1 depends on the value at 213.
[6] Kumar, P., Saini, R., Roy, P. P., & Dogra, D. P. (2017). A Bio-signal
time t, recurrent models such as LSTM and BLSTM help Based Framework To Secure Mobile Devices. Journal Of Network And
in preserving time dependencies and thus perform better than Computer Applications, 89, 62-71.
 8

[7] Marcel, S., & Milln, J. D. R. (2007). Person Authentication Using Power. Electroencephalography And Clinical Neurophysiology, 69(2),
Brainwaves (EEG) And Maximum A Posteriori Model Adaptation. IEEE 91-99.
Transactions On Pattern Analysis And Machine Intelligence, 29(4), 743 [30] Harmony, T., Marosi, E., de Len, A. E. D., Becker, J., & Fernndez,
- 752. T. (1990). Effect Of Sex, Psychosocial Disadvantages And Biological
[8] Wang, L., Zhang, X., Zhong, X., & Zhang, Y. (2013). Analysis And Risk Factors On EEG Maturation. Electroencephalography And Clinical
Classification Of Speech Imagery EEG For BCI. Biomedical Signal Neurophysiology, 75(6), 482-491.
Processing And Control, 8(6), 901-908. [31] Daz De Len, A. E., Harmony, T., Marosi, E., Becker, J., & Alvarez,
[9] Liao, L. D., Chen, C. Y., Wang, I. J., Chen, S. F., Li, S. Y., Chen, A. (1988). Effect Of Different Factors On EEG Spectral Parameters.
B. W., & Lin, C. T. (2012). Gaming Control Using A Wearable And International Journal Of Neuroscience, 43(1-2), 123-131.
Wireless EEG-based Brain-computer Interface Device With Novel Dry [32] Bennys, K., Rondouin, G., Vergnes, C., & Touchon, J. (2001). Diagnostic
Foam-based Sensors. Journal Of Neuroengineering And Rehabilitation, Value Of Quantitative EEG In Alzheimers Disease. Neurophysiologie
9(1), 5-16. Clinique/Clinical Neurophysiology, 31(3), 153-160.
[10] Franke, K., Ziegler, G., Klppel, S., Gaser, C., & Alzheimer’s Disease [33] Bonanni L, Thomas A, Tiraboschi P, Perfetti B, Varanese S, Onofrj
Neuroimaging Initiative. (2010). Estimating The Age Of Healthy Sub- M.(2008). EEG Comparisons In Early Alzheimers Disease, Dementia
jects From T1-weighted MRI Scans Using Kernel Methods: Exploring With Lewy Bodies And Parkinsons Disease With Dementia Patients
The Influence Of Various Parameters. Neuroimage, 50(3), 883-892. With A 2-year Follow-up. Brain: A Journal Of Neurology, 131, 690-
[11] Tipping, M. (2003). U.S. Patent No. 6,633,857. Washington, DC: U.S. 705.
Patent and Trademark Office. [34] Dauwels J, Srinivasan K, Reddy MR, Musha T, Vialatte FB,
[12] Gil Levi and Tal Hassner. Age And Gender Classification Using Convo- Latchoumane C, Jeong J, Cichocki A.(2011). Slowing And Loss Of
lutional Neural Networks. IEEE Conference On Computer Vision And Complexity In Alzheimers EEG: Two Sides Of The Same Coin?
Pattern Recognition (CVPR), Boston, 2015. International Journal Of Alzheimers Disease, 1-10.
[13] Lapuschkin, S., Binder, A., Mller, K. R., & Samek, W. (2017, August). [35] Cahill, L. (2006). Why Sex Matters For Neuroscience. Nature Reviews
Understanding And Comparing Deep Neural Networks For Age And Neuroscience, 7(6), 477-484.
Gender Classification. IEEE Conference On Computer Vision And [36] Guadalupe, T. et al.(2016). Human Subcortical Brain Asymmetries
Pattern Recognition, 1629-1638. In 15,847 People Worldwide Reveal Effects Of Age And Sex. Brain
[14] Mansanet, J., Albiol, A., & Paredes, R. (2016). Local Deep Neural Imaging And Behavaviour, 11(5), 1497-1514.
Networks For Gender Recognition. Pattern Recognition Letters, 70, 80- [37] Joel, D. et al.(2015). Sex Beyond The Genitalia: The Human Brain
86. Mosaic. Proceedings Of The National Academy Of Sciences 112(50),
[15] Liu, X., Li, J., Hu, C., & Pan, J. S. (2017, June). Deep Convolutional 15468-15473.
Neural Networks-based Age And Gender Classification With Facial [38] Arns, Martijn, Gerard Bruder, Ulrich Hegerl, Chris Spooner, Donna
Images. IEEE International Conference On Electronics Instrumentation M. Palmer, Amit Etkin, Kamran Fallahpour, Justine M. Gatt, Laurence
& Information Systems (EIIS), 1-4. Hirshberg, and Evian Gordon (2016). EEG Alpha Asymmetry As A
[16] C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, Gender-specific Predictor Of Outcome To Acute Treatment With Dif-
D. Erhan, V. Vanhoucke, and A. Rabinovich (2015). Going Deeper ferent Antidepressant Medications In The Randomized ISPOT-D Study.
With Convolutions. IEEE Conference On Computer Vision And Pattern Clinical Neurophysiology, 127(1), 509-519.
Recognition (CVPR), 1-9. [39] https://en.wikipedia.org/wiki/Bidirectional recurrent neural networks
[40] Khurana, V., Kumar, P., Saini, R. and Roy, P.P. (2018). EEG Based
[17] Y. Jia, E. Shelhamer, J. Donahue, S. Karayev, J. Long, R. Gir- shick,
Word Familiarity Using Features And Frequency Bands Combination.
S. Guadarrama, and T. Darrell (2014). Caffe: Convolutional Architec-
Cognitive Systems Research, 49, 33-48.
ture For Fast Feature Embedding. ACM International Conference On
[41] Kaur, B., Singh, D. and Roy, P.P. (2018). Age And Gender Classification
Multimedia, 675-678.
Using Brain-computer Interface, Neural Computing And Applications ,
[18] K. Simonyan and A. Zisserman (2014). Very Deep Convolutional
1-14.
Networks For Large-scale Image Recognition. International Conference
[42] Levi, G., Hassner, T. (2015). Age And Gender Classification Using
On Learning Representations.
Convolutional Neural Networks. IEEE Conference On Computer Vision
[19] http://www.openu.ac.il/home/hassner/Adience/data.html.
And Pattern Recognition Workshops, 34-4.
[20] M. Soleymani, S. Asghari-Esfeden, Y. Fu and M. Pantic (2016). Analysis [43] Ruigrok, Amber NV, Gholamreza Salimi-Khorshidi, Meng-Chuan Lai,
Of EEG Signals And Facial Expressions For Continuous Emotion Simon Baron-Cohen, Michael V. Lombardo, Roger J. Tait, and John
Detection, IEEE Transactions On Affective Computing, 7(1), 17-28. Suckling (2014). A Meta-analysis Of Sex Differences In Human Brain
[21] Davidson, P. R., Jones, R. D., & Peiris, M. T. R. (2005). Detecting Structure. Neuroscience & Biobehavioral Reviews, 39, 34-50.
Behavioural Microsleeps Using EEG And LSTM Recurrent Neural [44] Clarke, A. R., Barry, R. J., McCarthy, R., & Selikowitz, M. (2001). Age
Networks. IEEE International Conference On Engineering In Medicine and sex effects in the EEG: development of the normal child. Clinical
And Biology Society . neurophysiology, 112(5), 806-814.
[22] Ni, Z., Yuksel, A. C., Ni, X., Mandel, M. I., & Xie, L. (2017). Confused [45] Gupta, A., Sahu, H., Nanecha, N., Kumar, P., Roy, P. P., Chang, V.
Or Not Confused?: Disentangling Brain Activity From EEG Data Using (2018). Enhancing Text Using Emotion Detected from EEG Signals.
Bidirectional LSTM Recurrent Neural Networks. ACM International Journal of Grid Computing, 1-16.
Conference On Bioinformatics, Computational Biology, And Health [46] Chang, V. (2014). Cloud Computing for brain segmentationa perspective
Informatics, 241-246. from the technology and evaluations. International Journal of Big Data
[23] Golmohammadi, M., Ziyabari, S., Shah, V., de Diego, S. L., Obeid, I., & Intelligence, 1(4), 192-204.
Picone, J. (2017). Deep Architectures For Automated Seizure Detection [47] Karyotis, C., Doctor, F., Iqbal, R., James, A., Chang, V. (2018). A fuzzy
In Scalp EEGs. ArXiv ArXiv:1712.09776. computational model of emotion for cloud based sentiment analysis.
[24] Hasib, M. M., Nayak, T., & Huang, Y. (2018). A Hierarchical LSTM Information Sciences, 433, 448-463.
Model With Attention For Modeling EEG Non-stationarity For Human [48] Chang, V. (2018). An overview, examples, and impacts offered by
Decision Prediction. IEEE International Conference On Biomedical & Emerging Services and Analytics in Cloud Computing virtual reality.
Health Informatics (BHI), 104-107. Neural Computing and Applications, 29(5), 1243-1256.
[25] Schmidhuber, J., & Hochreiter, S. (1997). Long Short-term Memory. [49] Kumar, S., Yadava, M., & Roy, P. P. (2018). Fusion of EEG Response
Neural Computation, 9(8), 1735-1780. and Sentiment Analysis of Products Review to Predict Customer Satis-
[26] Putten, Michel JAM, Sebastian Olbrich, and Martijn Arns. (2018). faction. Information Fusion.
Predicting Sex From Brain Rhythms With Deep Learning. Scientific [50] Khurana, V., Kumar, P., Saini, R., & Roy, P. P. (2018). EEG based word
Reports, 8(1), 3069-3076. familiarity using features and frequency bands combination. Cognitive
[27] Yadava, M., Kumar, P., Saini, R., Roy, P. P., & Dogra, D. P. (2017). Systems Research, 49, 33-48.
Analysis Of EEG Signals And Its Application To Neuromarketing. [51] Yadava, M., Kumar, P., Saini, R., Roy, P. P., & Dogra, D. P. (2017).
Multimedia Tools And Applications, 76(18), 19087-19111. Analysis of EEG signals and its application to neuromarketing. Multi-
[28] Matouek, M., & Petersn, I. (1973). Automatic Evaluation Of EEG media Tools and Applications, 76(18), 19087-19111.
Background Activity By Means Of Age-dependent EEG Quotients. [52] Sharma, S., Chang, V., Tim, U., Wong, J., Gadia, S. (2016). Cloud-
Electroencephalography And Clinical Neurophysiology, 35(6), 603-612. based emerging services systems. International Journal of Information
[29] Gasser, T., Verleger, R., Bcher, P., & Sroka, L. (1988). Development Of Management, 1-12.
The EEG Of School-age Children And Adolescents. I. Analysis Of Band

View publication stats