A set of emotions drawn on some colourful yellow balls

Detecting Emotion [Using EEG & Fuzzy Logic]

By delving into brain-computer interfaces (BCI) and EEG research, talented students from the Myelin group at the University of Michigan embarked on an intriguing project: detecting human emotion using neurological signals. 

Using a customized video game, Mentalab’s cutting-edge mobile EEG headset, and trained machine learning (ML) models, they successfully collected data from participants and achieved a 94.7% success rate in anger detection.

In this article, we look at their approach, and explore the potential of fuzzy logic as a crucial component in accurately tracking and classifying emotions.

Classifying Emotions Using Neuroimaging

Understanding and detecting emotions poses a significant challenge. Neuronal interactions associated with emotion are intricate and inherently weak (Esslen et.al., 2004). This makes understanding responses to stimuli difficult.

The neuroimaging community have made little progress in effectively classifying emotions (Bagherzadeh et.al., 2019). However, students at the University of Michigan took on this challenge and devised a new method to induce and quantify emotions using fuzzy logic.

Limitations in Emotion Induction

Scientists interested in affect tend to induce emotions using static images. For instance, the GAPED dataset aims to evoke a range of emotions on an arousal-valence scale and assign them to over 760 images (Dan et al., 2011). 

Valence quantifies the amount of pleasure someone derives from a stimulus, while arousal quantifies how stimulated someone is; from excitement to lethargy (Du et.al, 2020).

The Myelin group discovered that many of the GAPED images elicit emotions lying on the anger-boredom spectrum (Athavipach et.al.,2019). This suggested that anger and boredom were easier to induce than other emotions. As such, they decided to focus on anger induction. 

Interested in EEG?

Video Games Evoke Anger

The students drew inspiration from discussions with industry experts like Alex Milenkovic, CEO of Curia, who has extensive experience collecting EEG data from e-sport teams.

Unlike static stimuli, such as images, video games provide a continuous emotional response. What is more, because players are invested (and addicted) to gameplay, they evoke a clearer and stronger reaction that replicates across trials.

This means that researchers can generate a larger signal-to-noise ratio using video games than they could using static images (which is good!).

Inspired, the students developed “Astrobird,” a modified version of the world-renowned game “Flappy Bird,” known for inducing rage in players.

Screenshots of the Astrobird game environment. On the right is a grame over screen.
Astrobird. Left shows what the users saw during gameplay. Right shows the “Game Over” screen.

The Myelin group incentivized good performance by punishing low scorers.

Mentalab Explore Imaged the Brain 

The students used Mentalab’s state-of-the-art mobile EEG headset to collect EEG data during gameplay. They considered 15, five-minute trials.

The Myelin group using the Explore system to measure emotion responses.
The Myelin group putting on the headset and setting up the Mentalab desktop application before recording EEG data.

Mentalab Explore collected data while participants played. As expected, gameplay produced clearer emotional responses than photos.

The students isolated events in four-second intervals. Either a player was in a neutral state, or had lost the game.

EEG signal comparing a random event to an end event in the game
EEG data during a trial, distiguishing reactions to neutral events (left) and game over events. Here, the y-axis is a unitless signal intensity, in which all values are divided by the maximum.

The team decided to use fuzzy logic to classify their data.

Fuzzy logic is a way of modelling data that uses logical reasoning. It considers the trutheness of a variable; it accepts that something may not be 100% true, nor 100% false. It also admits of outliers, noise, and vagueness (Gu et al., 2022).

In contrast, other classification methods are often binary, labelling variables as either true or false.

The Myelin group trained and compared eight classifiers for performance. A multilayer perceptron (MLP) and a support vector machine (SVM) provided the best results. They achieved an impressive 94.7% accuracy in anger detection.

Mobile EEG Classifies Emotion

Applying fuzzy logic to classify EEG data in real-time is not only easier than binary logic, but also paves the way for more innovative studies of the human brain.

By combining the portability and ease of use of Mentalab’s mobile EEG headset with the power of fuzzy logic, the Myelin group students were able to achieve remarkable success in detecting emotion.

The group aims to design more experiments that encompass a broader range of mental states. In essence, they hope their research could be used to detect neurodevelopmental disorders, such as ADHD, that often go undiagnosed. 

New mobile hardware, such as Mentalab’s Explore Pro system, play a key role in developing such applications.

If you are part of a student club, contact Mentalab to inquire about our student club support program which made this project possible


Athavipach, C., Pan-Ngum, S., & Israsena, P. (2019). A Wearable In-Ear EEG Device for Emotion Monitoring. Sensors (Basel, Switzerland), 19(18), 4014. https://doi.org/10.3390/s19184014

Bagherzadeh, S., Maghooli, K., Farhadi, J., & Zangeneh Soroush, M. (2018). Emotion recognition from physiological signals using parallel stacked Autoencoders. Neurophysiology, 50(6), 428-435. https://doi.org/10.1007/s11062-019-09775-y

Dan-Glauser, E. S., & Scherer, K. R. (2011). The Geneva affective picture database (GAPED): A new 730-picture database focusing on valence and normative significance. Behavior Research Methods, 43(2), 468-477. https://doi.org/10.3758/s13428-011-0064-1 

Du, N., Zhou, F., Pulver, E., Tilbury, D., Robert, L., Pradhan, A., & Yang, J. (2020). Examining the effects of emotional valence and arousal on takeover performance in conditionally automated driving. SSRN Electronic Journal, 112(0968090), 78-87. https://doi.org/10.2139/ssrn.3518015 

Esslen, M., Pascual-Marqui, R., Hell, D., Kochi, K., & Lehmann, D. (2004). Brain areas and time course of emotional processing. NeuroImage, 21(4), 1189-1203. https://doi.org/10.1016/j.neuroimage.2003.10.001

Gu, X., Han, J., Shen, Q., & Angelov, P. P. (2022). Autonomous learning for fuzzy systems: a review. In Artificial Intelligence Review (Vol. 56, Issue 8, pp. 7549–7595). Springer Science and Business Media LLC. https://doi.org/10.1007/s10462-022-10355-6