Estimation of Mental Workload, Essay Example

Estimation of mental workload by EEG signal analysis and machine learning for BCI

Abstract

Mental workload estimation from (EEG) offers a sensitive mechanism to in relation to the Human-computer interaction (HCI) connecting it to the user. In order to create a system that can work properly in the complexity of such a real world, there is essential need to achieve changes within a specific context for instance mood. Accordingly, Brain-computer interface (BCI) helps in controlling of external devices only with electrical activity of the human brain. Various studies have proposed approaches that can help improve the system. Nevertheless, it is crucial to test algorithms using a standard BCI signal either from internet repositories available or with the help of expert users. Due to inadequateness of comprehensive researches that have been conducted in the past, this study will test extreme learning machine (ELM) will involve at least 6 novel users signals in order to enable comparison with standard classification algorithms. From the hypothesis, results are expected to show that ELM is the most suitable criteria to help in classification of Electroencephalogram signals from novice users.

However, to understand the of state-of-the-art EEG based devise, a favorable experimental protocol that can control the intended contextual outcome, while novice users are solving tasks that have a minimum of two workload levels will be devised for a more reliable outcome. The study recorded personal recordings and physiology of the 6 novice users who participated in the study in order to ascertain the validity of the protocol. Use of either frequency or time or both domains tested the effectiveness of various subject-specific mental workload classifiers in order to generalize the contexts. Consequently, the study depicts the ability of classifiers to change with the contexts that are affective, although the outcome suffers independence of the involved domain. Nevertheless, a good solution is the cross context training in that allows extraction of feature domains in all involved domain varieties, which have unrelated task variations to signal characteristics. . The study findings are expected to help generate comprehensive findings estimated mental workload by analysis of machine learning for BCI and EEG signals.

Introduction

Background to project

This project is mainly focused on brain-computer interface (BCI) which studies the communication between the human brain and an external device such as a computer.

BCI has a long history and it has been researched since 1970 and has developed significantly in the past decade [1]. BCI has proved to be an important tool in different areas. One useful application of BCI is in helping disabled people to do daily life tasks just by using the activity in their brain like moving their wheelchairs or operating their prosthetic body parts using their brain only. One of the many companies which have been working on the application of BCI in the past decade is called Emotiv. Emotiv is a high-tech company which has been developing neuro headsets, software and SDK for over a decade. The Emotiv Epoc+ headset used in this project was released in 2013 and it is a wireless headset that has 14-channels and can record electroencephalogram (EEG) signals with them.

Epoc+ transmits data through Bluetooth and the various sensors in all 14 channels can recognize facial expressions such as muscle movement, mental commands and also 6 different emotional and sub-conscious dimensions in real time like excitement, stress, boredom and interest [2]-Epoc+ is a headset that can be used for countless different BCI applications such as basic tasks like controlling mouse controls by just thinking the commands and closing some files by blinking, or  more advanced ones like controlling a fully functional 3D animated character in a video game, moving/rotating wheelchairs and changing the size, length and width of 3D objects on a computer screen all by just using brain activity signals (EEG).In this project, a simple BCI application has been researched and its results will be demonstrated.

This paper is organized in sections that can be followed systematically to enable quick understanding of the process. Section two will focus on the literature review expounding on what other previous studies have come out with and concluded. Thereafter, the third section which is the methodology chapter will focus on the description of the materials used and how they were used for the experiment. Subsequently, the fourth chapter will be the description of the experiment while the next chapter will discuss the results and findings of the findings. Finally, the last chapter the main concluding remarks that are in line with the study findings

Objectives

Basic Objectives

Technical report writing, a literature review on machine learning for brain-computer interface (BCI), design protocol for experiments, recording brain activity while someone is indifferent test conditions, signal processing to extract data for further modeling and analysis.

Advanced Objectives

Development of subject-specific models of behavior/response using a suitable machine learning methodology, develop cohort-based models and validate their performance against subject-specific, add more physiological markers along the BCI interface to enhance modeling performance

 Justification/Significance of the Study

Mental workload estimation from Electroencephalographic signals (EEC) offers a highly sensitive mechanism to help adapt the Human-computer interaction (HCI) to the user. The study aims to create a system that can work properly in the complexity of such a real world. There is essential need to achieve robustness against contextual changes for instance mood. Accordingly, Brain-computer interface (BCI) will enable in controlling of external devices only with electrical activity of the human brain.

The scope of the Study

The proposed study will focus on assessing the mental workload of 6 novice users who will be put under three different experiments in order to achieve a more reliable outcome. More importantly is the training of classifiers who will help in conducting the research study.

Limitations

The researcher anticipates facing various challenges while conducting the study. One of the challenges that the researcher anticipates is financial challenges. This is because the study aims at collecting information using EEGLab and Matlab that are quite expensive. The other limitation is the expansiveness of the study making it difficult to be studied at ago. The researcher will, therefore, sample out representative novice users to be able to cope with time and financial resource constraints.

Operational Definition of Terms

Mental Workload: defined as (perceived) relationship between the amount of mental processing capability and the amount required by the task (Hart and Staveland, 1988)

Brain-computer interface: A device that helps to connect a computer and human brain through electoral signals           

EEGLab: is a free open source EEC program that works in conjunction with Matlab as a built-in toolbox and it is used for analyzing and extracting the raw EEC data that has been recorded during the in-lab experiments.

Emotiv Epoc+: a headset that has been used saves the raw signals in EDF format and using EEGLab these data can be imported and analyzed.

Chapter 3: Materials and Methodology

The study was designed in a manner in which participants had carry out mind challenging tasks involving three different workload experiments while handling different frequency-domain and time-domain. EEC signals were used collected with EEGLab and Matlab in collecting EEC raw data. Subsequently, the data was saved in EDF format using Emotiv Epoc+.

Participants

Six novice users participants volunteered for the experiments. The age limit for those who volunteered were between 18 and 32 years old, all of them being right-handed. Furthermore, People with visual problems and neurological conditions were excluded from the procedures carried out in order to obtain fair results.

Materials and Equipment

EEGLab and Matlab were used in collecting EEC raw data. Subsequently, the data was saved in EDF format using Emotiv Epoc+. For the image processing experiment, an apple was used to prevent any distraction and let the participant to only focus on a single simple image for obtaining more reliable and accurate signals. Finally, Participants signed an informed consent before being installed with the sensors. Each experiment was conducted under a specific recorded time.

Workload tasks

Three experiments were conducted, starting with the blinking experiment than a mathematical equation experiment and finally the image processing experiment. The experiments will be further described in the next section.

Self-assessment data

To evaluate the overall impact of psychosocial stress induction, an STAI score was used computed with an ANOVA in three different levels the ?baseline?, ?after relaxation,? and ?after stress induction. Subsequently, a 2 (stress) × 2 was conducted to assess stress and workload manipulation.

Physiological data

The physiological response was analyzed specifically for heart rate and galvanic skin response. However, the galvanic skin response mean was processed before engaging the statistical methods for each experiment. Likewise, the EEG was filtered between 5 and 200Hz through applying a filter notch. This was done in order to reduce power line noise. Finally, the ANOVA data for all participants were squared (?²_p) to measure impact size. Are calculated as a measure of effect size.

EEGLAB

EEGLab is a free open source EEG program that works in conjunction with Matlab as a built-in toolbox and it is used for analyzing and extracting the raw EEG data that has been recorded during the in-lab experiments. Emotiv Epoc+ headset that has been used saves the raw signals in EDF format and using EEGLab these data can be imported and analyzed. .The next important step of the process was to locate the channels in the program. The channel locations can be found in a CED file, which is extracted from the EDF files generated by the Epoc+ headset. While running the program with the given file, error messages kept occurring saying more columns than expected. This problem was ultimately solved by deleting the last column of the CED file, which was an unwanted and automatically generated column of ones. The final and working version of channel locations file is given below in figure 1. It consists of important data such as channel numbers 1 through 14 with their typical labels, Cartesian coordinates (x,y,z), radius and angles which are used to show the location of channels on a spherical head and are needed for plotting them.

Figure 1: Extracted channel locations from the EDF file

Baseline Removal

When channel locations have been loaded into the program baseline needs to be removed ?.it cleans things up quite a bit, Removes the mean of each data channel,

Independent Component Analysis (ICA) heading or part of others paragraph.

In addition, after its completion, clean EEG data is ready to be plotted and analyzed.

The figure 4 above is channel spectra and maps plot of some specified frequencies. It shows the location of activity taking place in the brain for each frequency. As this is a plot of a blinking experiment, it can be seen that most of the activity is taking place in the frontal lobe of the brain as that is the part responsible for functions such as facial expressions and blinking.

Filtering Artefacts

Artefacts are external unwanted signals that may cause inefficiency in results of signal processing and analysis if they are not filtered appropriately. They can be extraneous noise or all different kinds of signals such as electrocardiogram (ECG) which is recorded from heart rhythm and electromyogram (EMG) that detects electrical activity of muscles. In the case of experiments in this project, the biggest artefacts were EMG signals caused by blinking in second and third experiments where blinking was not the main interest. The process of filtering these artefacts is explained below.

The following figure shows a plot of recorded EEG signals of all channels and this plot of scroll data in the EEGLab toolbox has been used to remove artefacts such as unwanted facial muscle movement. For example, the image below shows the brain activity of a participant who is solving a mathematical equation (second experiment) and as the focus of this experiment is problem-solving and attention of a human, therefore, any unwanted blinking in the process is counted as artefacts and needs to be removed from the signals. To do so, in EEGLab, it is possible to highlight the artefacts or unwanted frequency ranges and by clicking ?reject’ they will be filtered from the original signals.

Even though most of the artefacts have been removed manually or automatically, it has to be said that it is impossible to remove all artefacts. This is because some of them are very negligible or they are merged with EEG signals in a way that their elimination will risk removing relevant information of the main signals under investigation.

Logistic regression is a statistical method for analyzing a dataset in which there are one or more independent variables that determine an outcome. The outcome is measured with a dichotomous variable (in which there are only two possible outcomes).

Logistic regression is a tool for modeling a categorical variable in terms of other variables. It gives you ways to find out how changes in each of the ?other? variables affect the odds of different outcomes in the first variable. The output is easy to interpret.

Neural networks are a set of methods to let a computer try to learn from examples in ways that vaguely resemble how humans learn about things. It may result in models that are good predictors, but they are usually much vaguer than those from logistic regression are.

Experiments

Three different experiments have been designed to be conducted on five participants. Each of these experiments is focused on analyzing a different part of the brain and they are explained in detail in this section.

Blinking Experiment

The first experiment designed for this project is a blinking experiment in which the electrical activity of the brain is recorded with Emotiv Epoc+ headset while blinking by placing the electrodes on the scalp. A 5-second time interval has been determined for blinking with a 60-second time limit for each test. The tests have been repeated several times and three of the most accurate results have been saved for later analysis. Accurate results mean the ones in which the participant has not blinked in between the specified 5-second intervals, there has been no other artefacts in the signals such as unwanted muscle movement, and where all 14 channels of the headset have good connections with the scalp. Moreover, this experiment is designed to investigate one of the functionalities of the frontal lobe of the brain, which is muscle movement as shown in figure 6.

The standard sampling rate used is 128 Hz per second.

Classical EEG rhythms used: ? (1–4 Hz), ? (4–8 Hz), ? (8–12 Hz), ? (12–30 Hz), ? (30–47 Hz), and high ? (53–90 Hz).

Mathematical Equation Experiment

The second experiment is focused on problem-solving and the participant is asked to solve a mathematical equation with a time limit of 30 seconds while their brain activity is recorded. They are asked to think through it slowly as imagining each of the steps in solving it so that the brain is focused mostly on problem-solving which is functionality in the frontal lobe and it can be analyzed. The experiment is conducted multiple times and two of the most reliable tests with least artefacts are saved for analysis. In this experiment even though the equation stays the same throughout the trials but the main interest has been put on the participant’s speed in solving the equation as it gets easier for them after the first time, and fatigue and tiredness have not been considered as it could devalue the main purpose. The following mathematical equation has been for this experiment.

The reason why this medium difficulty equation has been used is that if the equation was too complex. Again, some participants might not have been able to solve it in the 30-second time limit and on the other hand, if it was an easy problem then it could be solved in just a few seconds, which is not desirable. Therefore, this problem with addition, subtraction, and multiplication have been used so that participants who were mostly college students could be able to solve it in or slightly before the required time limit so there is consistency in the duration of experiments.

Image Processing Experiment

Third and the last experiment is designed to monitor the functionality of vision, which is in the occipital lobe of the human brain. Participants have to concentrate and look at the image of a red apple as shown in the figure below for 30 seconds each time and several tests are conducted so two of the most accurate ones can be selected for analysis. Same as the previous experiment, fatigue, and tiredness has been ignored in this scenario in order to focus on the parts, which need to be analyzed ? vision, and attention.

The reason why this fully red simple image of an apple with a white background has been used is to prevent any distraction and let the participant focus only on a single simple image for obtaining more reliable and accurate signals.

Chapter 4: Description

This chapter describes how the machine learning approach is applied to EEG signals so that the computer can be able to predict human actions by just analyzing the brain activity and this can be done through a special type of training and modeling.

EEG signal processing

Our protocol aimed at estimating the level of mental workload of the participants from signals data captured. Accordingly, a machine learning approach was introduced on basis of state-of-the-art algorithms created for a brain-computer interface. EEG signals were first pre-processed and subdivided into trials.

Machine learning approach to workload level classification from EEG signals

The machine  learning approach is classified into two levels set which include the training set and testing set. At the top, which is the training set, it aims at identifying required frequency bands and channels, for instance, the spectral filters and spatial filters. In our experiment, this will be achieved through the use of filter Bank based on the CSP and REFSF approaches. Subsequently, at the bottom is the testing set also known as the modeling set that will be applying the spectral and spatial filters in order to estimate the workload level from the EEG trial.

EEG processing and segmentation

At the beginning of the experiment, all signals from eye movements (EOG) were removed through cleaning through an automatic method proposed in Schlögl et al. (2007). Subsequently, all EEG signals from the various tasks were segmented into 60 different EEG trials. For example, it was modeled in such a way that every EEG trial would be represented by a letter appearance. To be precise, the EEG trials were defined within the 2 s from its onset appearance. Following this sequence, the total number of trials per task resulted in 60 EEG. Therefore, it meant 720 trials per workload level because the experiment was carried out in two levels of contextual changes, for instance, 360 trials per each condition.

In order to prevent any interfering effect that may result from Event-related potentials (ERP), trials that corresponded to target letters were not recorded. This trend usually arises as a result of trials that end up in target identification. Consequently, the trials recorded per context condition reduced to 270 trials per condition.

Machine learning algorithms

Our experiment based its observations on two types of neurophysiological information to help estimate workload levels from the EEG transmitted signals. These included the oscillatory activity and event-related potentials which according to Brouwer (2012) are effective for such a task. Consequently, the state-of-art signal processing pipelines were created to enable estimation of these two types of information at the individual level and a combined level.

Oscillatory activity

This activity was crucial in the classification of activity into low mental workload vs. high mental workload in the EEG signals. Furthermore, a variant of the Filter Bank Common Spatial Patterns (FBCSP) algorithms was put to use to help analyze optimal spatial and spectral features. According to various studies, FBCSP is the most efficient algorithms that can be used to extract spatial-spectral features while analyzing EEG signals.

The first step of this approach is the training phase that consists of choosing the most useful frequency band and EEG channels by use of signals from high and low workload conditions. Firstly, each training EEG trial was filtered to produce multiple frequency bands with the help of a bank of band-pass filters. Subsequently, the band-pass filtered EEG trials of each of these frequency bands to optimize spatial filters such as linear combinations of the original EEG channels.

Classical EEG rhythms used: ? (1–4 Hz), ? (4–8 Hz), ? (8–12 Hz), ? (12–30 Hz), ? (30–47 Hz), and high ? (53–90 Hz).

The common spatial pattern (CSP) algorithm used to optimize these spatial filters helps to establish an optimal channel combination. As a result, the power of the outcome spatiality filtered signals is fully discriminant between the two conditions. On the other hand, the testing phase whose main objective is to predict workload level of a given trial, the EEG signals are filtered is selected with 18 particular pairs of spatial and spectral filters. Finally, the output power is computing and challenged as input to the previously trained LDA classifier whose resulting signals would present the workload level as either high or low.

Chapter 5: Findings and Discussion

The results that were compiled because of the evaluation process described in the methodology are presented in a separate attachment labeled in Appendix A. The individualized data and statistics pertaining to each of the studies are also included in a separate attachment. The estimation of mental workload by EEG signal analysis generated significant results pertaining to its mental workload. The findings are consistent with the findings in other research models covered in the literature review. The patterns consistent with mental work overload are covered in the experiment and so the effectiveness of the analysis process is demonstrated.

An overcomplete spectral regression (OSR) presents consistent patterns of information relating to the eyes and mental overwork patterns of patients as they perform various tasks. From the Adaptive Mixture of Independent Component Analysis, the computer software is able to generate the patterns associated with the subject at each moment of time goes by. The resulting representation of this is shown in the form of peaks and troughs throughout the graph, which is then represented on a matrix consisting of a variety of other factors overlapped together through the sensory software.

The use of this analysis presents the future of brain-computer interface and there are countless possibilities for the different applications this type of technology can have. From the analysis of the findings, when the brain is under some type of stress or pressure, the overall performance of the brain-computer interface was not relevant. Therefore, the research opted to find a way of controlling effective context such as mood while using the system.

From the informed and consenting volunteers, the raw data from the experiment with the sample size of 20 trials from 6 volunteers indicate consistent patterns in the starting and stopping of each blink of the eye throughout the first section of the experiment to be explained technically in detail shortly. The second and third experiments involving the solving of a math equation and the looking at images simultaneously were conducted afterward. The following is an example of some of the results from this study

To the right is an example of how the information shows up and is presented from the Xavier Testbench software, which is presented in the appendix.

All of the raw information generated throughout the experimental study has been recorded in the Xavier Testbench software, which makes then provides it in an understandable form to the user. Indicated on the results of each person were consistent peaks and toughs now of each eye blinking in the first part of the experiment. The recorded information is presented in the appendix of this study. The EEG signals are captured through the software making them of value for the study. Much of the results from participant to participant have been consistent with each other in the demonstration of the blinking part of the experiment. This shows that the software is capable of consistently capturing, recording, storing, and presenting data that comes from the process of evaluating the different parts of the brain as it performs a variety of tasks. It not only affirms the effectiveness of the software but also has application to the medical context as well.

Discussion

This experiment has produced results consistent with the current hypothesis of the study, which seeks to capture the brain activity and reflect it to the EEG signals, which then through an external device enable performance of an activity. The outcome of the study is important as it can help design systems that can help physically disadvantaged persons to conduct activities through brain detection by use of the brain-computer interface. More importantly, through image capturing activities such as blinking of an eye, they may be able to control computer applications. These different ways this software can be applied makes it perhaps one of the more interesting aspects of this software.

There are different models that exist when it comes to the evaluation of brain interface technology. The success of this software in its ability to capture the patterns of brain waves can be applied in the neuroscience and help neuroscientists conducted further research that might prove of importance to the field of medicine.

Although theoretical neuroscience dates back to 1924 and modern neuroscience dates as far back as 1970, this is the first time professionals have been able to measure characteristics of the brain with this much detailed accuracy live as it is happening. It surpasses the standard brain scan, which would indicate photographically whether something exists, or not. This functions as a live detection system for any irregular or unusual activity and even as it may be associated with a particular substance.

This is interesting because of the application that it can have in a wide variety of fields. It can be used to figure out psychologically or psychiatrically what are the things that bring the most significant stressor overload from. Maybe some people are working too much and so they are experiencing a mental stress overload. Being able to identify these types of issues with patients has the significant value of being able to help guide people to treatment in the right direction. The frequency brands associated with each activity were consistent with most of the participants, which was the most surprising element of the study. The experiment shows that the software available in this area of study is the most advanced it has ever been. Each lobe is measured for and plotted accordingly for observation during the study, which is the main objective of the software and equipment. A couple of factors seemed to influence the outcomes of the consented volunteers in this experiment.

With regards to the frequency brands associated with each of the different levels of stressors and the different portions of the brain, the software was able to capture some block transition of action as it is happening. The first trial runs of the experiment were difficult because it did not seem like the connection between the software and hardware was connecting properly. It could be because the software has not been fully developed in the strength of its detection or it could be due to some error on the user side of the handling of the software. Nonetheless, there was some capture of block-to-block pulsating movements recorded in the software and they were consistent with essentially all of the participants.

It is clear from the study that the Xavier Testbench study is just as capable as other types of EEG analysis software. Furthermore, it seems that the Xavier Testbench only differs from other soft wares of the type moderately in the user interface and not much else. The software seems to be functional and competitive with other types of soft wares in the field. This has some indication that the software may have its specialized uses in the field later down the line depending on the contexts. It most likely depends on the preferences of the professional and the needs of their projects in particular when evaluating which soft wares to use.

Being able to utilize this software professional and with meaning requires years of experience and specialized training. It also comes with a drive and a passion. When there is an objective to be worked for, such as a psychiatrist looking to improve the lives of their patients by understanding perhaps what part of their brain being overworked that passion will drive professionals to make further discovery. Although this experiment is backed by strong research in the subject and some training with the software, there is, still a significant amount of experience that this project is lacking to drive the level of technical detail of this software forward. Although this experiment focuses on the simple stress of working on a simple math problem or the task of blinking repeatedly for during controlled intervals, later experiments can focus on the reach to psychological or therapeutic treatments. All of these can be used in the future, which is perhaps the greatest value that can be added from this software.

Some of the focus has been on the ability of EEG analysis to capture the function of different parts of the brain during the mental stressors of real life, which typically take place on the go and are immeasurable. There are some limitations to the functionality of the software, which requires connection to the equipment for proper usage. This instantly presents the need for technology that is capable of performing the same functions wirelessly without the connection. This would make it so that this type of information is capable of being obtained from test subjects on the go, as they are free to go about their regular lives. Perhaps this may be the future of this field and technology as it has already advanced so much since its modern origin in the 1970s.

There is a significant amount of consideration into the expectations from a study in this category of computer science for several reasons. As a result, when the authors Kothe and Makeig investigate new and current tools for the use of EEG data, their analysis of the process is supportive but indicates there is a significant amount of improvement that can be made. Different models are put to comparison and their conclusion indicates a successful extraction of information from the software they used in that particular study known as BCILAB. Their study indicates their process was a success and so this research indicates that the process for this study has also been a success in that the information has also been extracted from the Xavier Testbench software system. The elements in that established scholarly research on the subject are also present in this research study, which leads to the confident conclusion that this research study has been a success.

From the information gathered during this experimental study of a random sample of subjects, it seems that the capabilities of EEG analysis have expanded in the most recent year significantly. Not only does this help alleviate some of the concerns of neurology in the past, it also presents the opportunity for a wide variety of professionals to utilize this software to their advantage. In addition, this software can be of great advantage to many organizations in increasing the quantity of output. For example, in the telecommunication industry, most of the activities can be set to be operated automatically by detection of brain activity.

Conclusion

In conclusion, it appears that the application of the brain-computer interface in connection with an external device such as a computer can play an integral part in improving the life of the physically disabled persons. For instance, through detection of brain activity, they can be able to control their wheelchairs. More important is that with accurate analysis of brain activity, further research studies can help revolutionize the medical industry. The results yielded consistent patterns associated with the various tasks the consented volunteers were asked to perform. The results indicate that the use of EEG is reliable and should be considered for the versatility and applicability it has to a wide variety of fields ranging from medicine to psychology. This software as demonstrated in this research project indicates that it is able to bring in a new generation of brain interface technology with a level of detail and accuracy greater than ever before.

References

[1] Jacques J. Vidal. 1973. Toward Brain-Computer Communication

[2] Emotiv Epoc+.[November 2017]. Available from: www.emotiv.com

[3] C Viddauree. C Channel. K Muller. B Blankertz. 2011.Machine-Learning-Based Coadaptive Calibration for Brain-computer interfaces. [November 2017]. Available from: www.mitpressjournals.org/doi/pdf/10.1162/NECO_a_00089

[4] R Posthuma. A Dagher. 2006. Cerebral Cortex. volume 16:10. Pg.1508-1521.

[5] Miguel Ortiz. 2012. A Brief History of Biosignal -Driven Art. [November 2017]. Available from: www.econtact.ca/14_2/ortiz_biofeedback.html

[6] Terrace J.Sejnowski. 2007. Toward Brain-Computer Interfacing.

[7] B Blankertz. R Tomioka. S Lemm. M Kawanabe. KR Muller. 2008. Optimizing Spatial Filters for Robust EEG Single-Trial Analysis. [November 2017]. Available from: www.ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=4408441&tag=1

[8] JM Vesin. T Ebrahimi. K Diserens. 2008. An efficient brain-computer interface for disabled subjects. [November 2017]. Available from: www.sciencedirect.com/science/article/pii/S0165027007001094

[9] KR Muller. M Tangermann. G Downhege. M Krauledat. G Curio. B Blankertz. 2008. Machine learning for real-time single-trial EEG-analysis from brain-computer interfacing to mental state monitoring. [November 2017]. Available from: www.sciencedirect.com/science/article/pii/S0165027007004657

[10] Live. love and learn psychology. [November 2017]. Available from: www.liveloveandlearnpsychology.com/brain.php

[11] Virtual reality place. [November 2017]. Available from: www.vrplace.de/produkt/emotiv-epoc-neuroheadset

[12] Electrode location on Epoc+ headset. [November 2017]. Available from: www.elemaya.it/XEmotivEpoc.htm