Intelligent human emotion recognition based on elephant herding optimization tuned support vector regression

doi:10.1016/j.bspc.2018.05.039

Biomedical Signal Processing and Control

Volume 45, August 2018, Pages 182-191

https://doi.org/10.1016/j.bspc.2018.05.039 Get rights and content

Highlights

•
Implement an intelligent human emotion recognition based on elephant herding optimization tuned support vector regression.
•
Discrete wavelet transform (DWT) was applied for features extraction electroencephalography signals.
•
Predicted three emotional scales as continuous variables including valence, dominance, and arousal.
•
Results of emotion regression by SVR classifier show that EHO can improve regression accuracy significantly.

Abstract

The ability to recognize emotional states of people surrounding us is an important portion of natural communication as emotions are fundamental factors in human decision handling, interaction, and cognitive procedure. The primary intention of this paper is to present an approach that uses electroencephalography (EEG) signals to recognize human emotions. This work targets emotional recognition in terms of three emotional scales; valence, arousal and dominance. EEG raw data were pre-processed to remove artifacts, discrete wavelet transform (DWT) was applied for features extraction. Moreover, support vector regression (SVR) is combined with Elephant herding optimization (EHO) to predict values of the three emotional scales as continuous variables. Multiple experiments are applied to evaluate prediction performance. EHO was applied in two stages of optimization. Firstly, to fine-tune regression parameters of the SVR. Secondly, to select the most relevant features extracted from all 40 EEG channels and eliminate ineffective and redundant features. To verify the proposed approach, results proved EHO-SVR ability to gain relatively enhanced performance measured by regression accuracy of 98.64%. Therefore, SVR is introduced in this paper as a better technique for predicting emotions as quantifiable continuous variables rather than classifying emotions into discrete emotional values.

Introduction

Human emotions are vital for human machine interaction (HMI) and affective computing. Some emotional examples are happiness, sadness, surprise and anger. Recently, there are many research efforts focusing on brain activity analysis, where subjects experience different emotional states. The ability to realize emotional states of people is an important part of natural communication. Currently, considering how much machines are involved in our lives, realizing human emotions by machines has been a research trend in brain computer interfacing (BCI) and advanced HMI. A reliable system for emotion recognition is a vital requirement to cope with the evolution of HMI. such system is required to grant good recognition performance, artifact tolerance, and adaptation to practical applications. In past years, a number of research efforts have been introduced in the field of emotion recognition [1].

With regards to EEG-based emotion recognition, various techniques have been proposed such as support vector regression (SVR), neural networks (NN), K-nearest neighbor (k-NN) and fuzzy systems. However, there have been significant findings proved that SVR is a superior regression technique in the field of emotion recognition based on EEG signals [2]. However, all regression techniques depend on a set of regression tuning parameters and kernel functions. According to what can be demonstrated from the experimental results in [3], regression performance can be significantly affected by such parameters. In [4], authors have proposed an emotion regression approach applying supervised learning and Bayesian classifier. Feature extraction stage employed fast Fourier transform. Then feature selection was achieved through the application of correlation coefficients.

Several studies [5], [6] have employed feature extraction, regression and classification. In this section, we review few research efforts which explore EEG feature extraction and regression problem in the domain of human emotion to detect emotion based on different classifiers. Brain emotional system is considered the substrate for emotion-related processes, EEG can expose intrinsic information on their task. Research on how EEG activity is related to emotions have created a lot of attention [7]. Furthermore, lots of studies have concentrated on the issue of asymmetric activation of the encephalic hemisphere. Also, Davidson et al. [8], stated that positive and negative emotions are related to relative left and right frontal EEG activities respectively. Baumgartner et al. [9], have pointed out that happy conditions are related to an extension to the left hemisphere, in contrast with unfavorable emotional conditions. In addition to research on asymmetrical activation of the encephalitic hemisphere, event-related potentials have also been utilized to study the association with EEG signals and human emotion.

The remainder of this paper is drawn as follows; literature review is introduced in Section 2. Section 3 presents a review of materials and methods used in this paper. In Section 4, research methodology is illustrated including EEG data, feature extraction, feature selection, emotion classification approach. The results of emotion detection are provided in Section 5 and at the end, Section 6 concludes the paper providing discussions on issues regarding the proposed approach and future work.

Section snippets

Literature review

The comparison among different studies in terms of machine learning techniques is not an easy task. Most studies comprise different preprocessing, features processing, experiment environments and other variations. However, such factors strongly affect the performance of a classification model. EEG data classification and regression have been the target of many research efforts under different machine learning techniques [10]. Some examples are K-Nearest neighbor (KNN), support vector machine

Materials and methods

Next paragraphs discuss the most vital techniques applied in this work and the targeted data set.

Proposed EHO-SVR approach

This section describes the proposed EHO-SVR approach for emotion regression using EEG signals. This approach is mainly based on optimizing regression accuracy of SVR by fine-tuning regression parameters in order to find best regression model of best performance. The proposed approach relied on RBF kernel as it can analyze higher dimensional data and it is a common approach for its good performance. RBF depends on two parameters and both are needed to be tuned for optimal discrimination which is

Experimental results and discussion

The experimental setup of dataset is explained in the following subsections. In this paper, DWT was employed where each signal frame generated a set of features. EEG signals features are characterized by their variation in values between individuals for the same feature. The proposed approach was employed to optimize regression of emotional scales from EEG signals. Experiment was held on data acquired from 5 subjects. DWT sub-bands were analyzed to a set of statistical features to generate

Conclusion and future work

This paper introduced a swarm-based technique for accurate regression of emotional states using SVR regressor. In this paper, we predicted three emotional scales as continuous variables: valence, dominance, and arousal. Five features are extracted for each signal band, for each channel and for each subject which are: Min, Max, Average, standard deviation, band entropy and the variance on time window of 4 seconds. The feature selection based on EHO is adapted and the SVR classifier parameters

References (43)

X.-W. Wang et al.
Emotional state classification from EEG data using machine learning approach
Neurocomputing
(2014)
H.J. Yoon et al.
EEG-based emotion estimation using Bayesian weighted-log-posterior function and perceptron convergence algorithm
Comput. Biol. Med.
(2013)
G. Chanel et al.
Short-term emotion assessment in a recall paradigm
Int. J. Hum. Comput Stud.
(2009)
Q. Zhang et al.
Analysis of positive and negative emotions in natural scene using brain activity and gist
Neurocomputing
(2009)
L. Zhang et al.
Intelligent facial emotion recognition using moth-firefly optimization
Knowl. Based Syst.
(2016)
J. Yang et al.
Channel selection and classification of electroencephalogram signals: an artificial neural network and genetic algorithm-based approach
Artif. Intell. Med.
(2012)
A. Ghaemi et al.
Automatic channel selection in EEG signals for classification of left or right hand movement in brain computer interfaces using improved binary gravitation search algorithm
Biomed. Signal Process. Control
(2017)
G.-G. Wang et al.
Elephant herding optimization
E.D. Übeyli
Wavelet/mixture of experts network structure for EEG signals classification
Expert Syst. Appl.
(2008)
M. Li et al.
Classification of epilepsy EEG signals using dwt-based envelope analysis and neural network ensemble
Biomed. Signal Process. Control
(2017)

A. Subasi et al.

EEG signal classification using pca, ica, lda and support vector machines

Expert Syst. Appl.

(2010)

P.C. Petrantonakis et al.

Emotion recognition from EEG using higher order crossings

IEEE Trans. Inf. Technol. Biomed.

(2010)

S.Y. Chung et al.

Affective classification using Bayesian classifier and supervised learning

A. Hamad et al.

A hybrid EEG signals classification approach based on grey wolf optimizer enhanced svms for epileptic detection

International Conference on Advanced Intelligent Systems and Informatics

(2017)

A. Hamad et al.

Feature extraction of epilepsy EEG using discrete wavelet transform

J.R. Estepp et al.

An assessment of non-stationarity in physiological cognitive state assessment using artificial neural networks

Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBC

(2011)

R.J. Davidson

Anterior electrophysiological asymmetries, emotion, and depression: conceptual and methodological conundrums

Psychophysiology

(1998)

J.B. Henriques et al.

Left frontal hypoactivation in depression

J. Abnormal Psychol.

(1991)

E.H. Houssein et al.

EEG signals classification for epileptic detection: a review

A.T. Sohaib et al.

Evaluating classifiers for emotion recognition using EEG

P. Rani et al.

An empirical study of machine learning techniques for affect recognition in human–robot interaction

Pattern Anal. Appl.

(2006)

Cited by (50)

Predicting clay compressibility for foundation design with high reliability and safety: A geotechnical engineering perspective using artificial neural network and five metaheuristic algorithms
2024, Reliability Engineering and System Safety
This study aims at predicting clay compressibility (clay compressibility) for foundation design with unprecedented reliability and safety. To achieve this goal, we set forth several scientific objectives to enhance the accuracy of clay compressibility predictions and ensure the robustness of our models. Firstly, we explored an updated dataset comprising diverse clay types, focusing on the intricate relationship between clay compressibility and essential soil properties. By incorporating particle size distribution, the ratio between liquid limit (w_LL) and plasticity index (PI), and the ratio between void ratio (e_LL) and wLL, our objective was to gain a comprehensive understanding of clay behavior, facilitating more accurate predictions. Secondly, we employed Artificial Neural Networks (ANNs) as our primary modeling tool. ANNs, with their capacity to discern complex patterns within data, formed the foundation of our predictive model. Our scientific objective was to harness the power of ANNs to capture subtle relationships among input variables, thereby ensuring a robust foundation for clay compressibility predictions. To further refine our predictions, our third objective was to optimize the ANN model. We utilized five distinct metaheuristic algorithms—Bird Swarm Algorithm (BSA), Dragonfly Optimization (DO), Elephant Herding Optimization (EHO), Jaya Algorithm (JA), and Whale Optimization Algorithm (WOA). Our aim was to fine-tune the ANN model's parameters using these algorithms, capitalizing on their strengths to achieve superior predictive performance. Additionally, our study aimed to pioneer the development and evaluation of hybrid models for forecasting clay compressibility. By integrating the ANN model with the metaheuristic algorithms (BSA-ANN, DO-ANN, EHO-ANN, JA-ANN, and WOA-ANN), our objective was to create hybrid frameworks that synergistically combined the strengths of both approaches. We rigorously evaluated these hybrid models, ensuring their efficacy and reliability in practical engineering applications. Crucially, we validated our models through practical engineering scenarios. For this purpose, we prepared 30 new soil samples mirroring real-world conditions. Our objective was to apply the developed models to these samples, affirming their effectiveness and reliability in actual foundation design applications. In pursuit of identifying the most accurate and reliable predictive model, our final objective focused on the BSA-ANN model. Remarkably, this model demonstrated an exceptional accuracy of 99.8%. Our objective was to establish the BSA-ANN model as the preferred choice for practical engineering applications, ensuring high reliability and safety in foundation design.
Fick's Law Algorithm: A physical law-based algorithm for numerical optimization
2023, Knowledge-Based Systems
Citation Excerpt :
In this subsection, an FLA assessment is done to test exploitative and exploratory abilities for all used benchmark. In general, optimization issue in the engineering design is a prominent study area, and literature uses many optimization methods to tackle this kind of problem [88–92]. This section evaluates the FLA algorithm’s performance in real-world engineering applications such as tension/compression spring design, welded beam construction, pressure vessel design, speed reducer design, and three-Bar truss design.
Recently, many metaheuristic optimization algorithms have been developed to address real-world issues. In this study, a new physics-based metaheuristic called Fick’s law optimization (FLA) is presented, in which Fick’s first rule of diffusion is utilized. According to Fick’s law of diffusion, molecules tend to diffuse from higher to lower concentration areas. Many experimental series are done to test FLA’s performance and ability in solving different optimization problems. Firstly, FLA is tested using twenty well-known benchmark functions and thirty CEC2017 test functions. Secondly, five real-world engineering problems are utilized to demonstrate the feasibility of the proposed FLA. The findings are compared with 12 well-known and powerful optimizers. A Wilcoxon rank-sum test is carried out to evaluate the comparable statistical performance of competing algorithms. Results prove that FLA achieves competitive and promising findings, a good convergence curve rate, and a good balance between exploration and exploitation. The source code is currently available for public from: https://se.mathworks.com/matlabcentral/fileexchange/121033-fick-s-law-algorithm-fla.
Honey Badger Algorithm: New metaheuristic algorithm for solving optimization problems
2022, Mathematics and Computers in Simulation
Citation Excerpt :
This makes these techniques highly inefficient in solving real optimization problems, leading to a growing interest in stochastic optimization techniques over the last two decades [31,45]. Often, most of the real-world optimization problems, in the area of engineering [17], wireless sensor networks [1], image processing [22], feature selection [27,44], tuning of machine learning parameters [18], bio-informatics [13] etc., are highly non-linear and non-convex due to inherent complex constraints and many design variables. Therefore, solving these types of optimization problems is complex because of many inherent local minima.
Recently, the numerical optimization field has attracted the research community to propose and develop various metaheuristic optimization algorithms. This paper presents a new metaheuristic optimization algorithm called Honey Badger Algorithm (HBA). The proposed algorithm is inspired from the intelligent foraging behavior of honey badger, to mathematically develop an efficient search strategy for solving optimization problems. The dynamic search behavior of honey badger with digging and honey finding approaches are formulated into exploration and exploitation phases in HBA. Moreover, with controlled randomization techniques, HBA maintains ample population diversity even towards the end of the search process. To assess the efficiency of HBA, 24 standard benchmark functions, CEC’17 test-suite, and four engineering design problems are solved. The solutions obtained using the HBA have been compared with ten well-known metaheuristic algorithms including Simulated annealing (SA), Particle Swarm Optimization (PSO), Covariance Matrix Adaptation Evolution Strategy (CMA-ES), Success-History based Adaptive Differential Evolution variants with linear population size reduction (L-SHADE), Moth-flame Optimization (MFO), Elephant Herding Optimization (EHO), Whale Optimization Algorithm (WOA), Grasshopper Optimization Algorithm (GOA), Thermal Exchange Optimization (TEO) and Harris hawks optimization (HHO). The experimental results, along with statistical analysis, reveal the effectiveness of HBA for solving optimization problems with complex search-space, as well as, its superiority in terms of convergence speed and exploration–exploitation balance, as compared to other methods used in this study. The source code of HBA is currently available for public at https://www.mathworks.com/matlabcentral/fileexchange/98204-honey-badger-algorithm.
Emotion recognition with EEG-based brain-computer interfaces: a systematic literature review
2024, Multimedia Tools and Applications
American zebra optimization algorithm for global optimization problems
2023, Scientific Reports
Boosting Kernel Search Optimizer with Slime Mould Foraging Behavior for Combined Economic Emission Dispatch Problems
2023, Journal of Bionic Engineering

View all citing articles on Scopus

¹: http://www.egyptscience.net.

View full text

Intelligent human emotion recognition based on elephant herding optimization tuned support vector regression

Highlights

Abstract

Introduction

Section snippets

Literature review

Materials and methods

Proposed EHO-SVR approach

Experimental results and discussion

Conclusion and future work

Neurocomputing

Comput. Biol. Med.

Int. J. Hum. Comput Stud.

Neurocomputing

Knowl. Based Syst.

Artif. Intell. Med.

Biomed. Signal Process. Control

Expert Syst. Appl.

Biomed. Signal Process. Control

Expert Syst. Appl.

Emotion recognition from EEG using higher order crossings

IEEE Trans. Inf. Technol. Biomed.

Affective classification using Bayesian classifier and supervised learning

A hybrid EEG signals classification approach based on grey wolf optimizer enhanced svms for epileptic detection

International Conference on Advanced Intelligent Systems and Informatics

Feature extraction of epilepsy EEG using discrete wavelet transform

An assessment of non-stationarity in physiological cognitive state assessment using artificial neural networks

Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBC

Anterior electrophysiological asymmetries, emotion, and depression: conceptual and methodological conundrums

Psychophysiology

Left frontal hypoactivation in depression

J. Abnormal Psychol.

EEG signals classification for epileptic detection: a review

Evaluating classifiers for emotion recognition using EEG

An empirical study of machine learning techniques for affect recognition in human–robot interaction

Pattern Anal. Appl.