JOURNAL OF CRITICAL REVIEWS
ISSN- 2394-5125 VOL 7, ISSUE 13, 2020
1707
ADAPTIVE FEATURE SELECTION BASED
LEARNING MODEL FOR EMOTION RECOGNITION
Mr.K. Anguraju
1
, Mr.N. Suresh Kumar
2
, Mr.S. Jerald Nirmal Kumar
3
, Mr.K. Anandhan
4
, Ms.P. Preethi
5
1
Assistant Professor, CSE, Kongunadu College of Engineering and Technology. E-mail: anguraju.k@gmail.com
2
Assistant Professor, SCSE, Galgotias University. E-mail: sureshkumar@galgotiasuniversity.edu.in
3
Assistant Professor, SCSE, Galgotias University. E-mail: jerald.kumar@galgotiasuniversity.edu.in
4
Assistant Professor, SCSE, Galgotias University. E-mail: anandhan.k@galgotiasuniversity.edu.in
5
Assistant Professor, CSE, Kongunadu College of Engineering and Technology.
E-mail: preethinest.1991@gmail.com
Received: 20.04.2020 Revised: 22.05.2020 Accepted: 17.06.2020
ABSTRACT: Human emotion recognition gives an extensive contribution towards human-computer interaction
with promising solutions for various Artificial Intelligence based problems. One major challenge is task
recognition with learning process to enhance performance while using Electroencephalogram (EEG) signals.
Here, two processes have been involved to enhance the classification accuracy. First, feature selection is
performed using proposed Adaptive Information Embedding (AIE) where the selected features are provided as
an input to classifier model. This AIE is used for reducing emotion signal strength with weighted combination of
contextual features. Secondly, spatial information is given to Convolutional Neural Networks (CNN) for
mapping the feature inputs and to classify the emotions of human more accurately. The simulation is done in
MATLAB 2018a environment. The proposed model shows better trade-off while comparing with existing
approaches. Accuracy is considered as metrics to validate the prediction factor. It gives an interpretable solution
for extensive EEG-based classification process.
KEYWORDS: Emotion recognition, Adaptive information embedding, Convolutional Neural networks, Feature
selection, Spatial-decoder.
© 2020 by Advance Scientific Research. This is an open-access article under the CC BY license
(http://creativecommons.org/licenses/by/4.0/) DOI: http://dx.doi.org/10.31838/jcr.07.13.267
I. INTRODUCTION
Human emotion shows higher impact in day to day activities which are related to activities like entertainment,
work and relaxation [1]. It gains the attention of huge investigators in relationship with physiological and
emotional activities. It is identified that positive emotional factors could provide pleasurable engagement and
gives beneficial advancements in human attitude and health. Moreover, it is related to various complications like
negative emotions, physical symptoms which causes adverse influence towards mental health and leads to crucial
psychological issues [2]. When information is explosive via social channels, it is extremely challenging to
project the emotional intimations. Now-a-days, learning based computation has gained huge demand for
reasonable utilization and deep knowledge for emotion recognition [3]. It is considered as a promising field of
research that has gained the attention of huge investigators in cross-circular regions ranging from neuro-science
to engineering. Emotions are delicately influenced due to various psychological and multiple external factors
with combination of space, time, experience and cultural background [4]. This increases the difficulties of
recognizing emotion in research field. Even though, huge efforts are made to explore mechanisms and techniques
for recognition of emotion, owing to intricate external patterns [5]. The effectual emotion recognition techniques
still needs huge demand in various technological applications.
Here, EEG is considered as a most preferred source for recognition of emotion due to its information richness
and finest temporal resolution. When compared to other approaches, EEG signals are considered to be more
effectual and authentic with superior unforgeability factors [6]. Various analyses have validated those
correlations among EEG signals and emotional states in various brain regions. However, with popularization and
development of wearable EEG devices and dry electrode approaches, on-line EEG signals are easily executed
and are more promising for diverse practical application with diverse tasks like disease detection, sleep stage