Hello,
I am currently working on a project with a colleague that uses EEG data to classify emotions using the circumplex model (i.e. valence/arousal). We plan to use the DEAP dataset for emotion calibration. However, one difference my colleague and I had was whether:
- Pre-recorded EEG data from the DEAP dataset can be directly used to train a classifier? OR
- Is it necessary to record participant's live EEG data, while being instructed to view items from the DEAP database to effectively categorize emotions with a classifier?
One issue I had is that pre-recorded EEG data from the DEAP database would not be as accurate for classification as having a group of participants view items from the DEAP database, while EEG activity is being recorded. However, my colleague suggests that recording raw EEG data from participants will be too time-consuming, and less effective. Does anyone familiar with EEG and emotion classification have any insights? Any suggestions are appreciated.
DEAP dataset
http://www.eecs.qmul.ac.uk/mmv/datasets/deap/readme.html
Luckily, you already have the DEAP dataset to work with! Have you tried a leave-one-out/bootstrapping approach where you develop a classifier from N-1 participants and use it to categorize the remaining participants data? The process can be reiterated in order to reduce sampling variance.
I think the bigger issue is what your intention is. Are you hoping that you might develop a model that will be able to generalize to new participants or are you hoping to use individual data to be able to predict subjective responses? It seems like using the DEAP classifier would be appropriate for the first and real-time EEG for the latter.
Let us know what you came up with!
Definitily there are many scope to construct a good model of emotion recognition. You may follow these papers to get the idea:
1. https://doi.org/10.1155/2014/627892
2. Yu Chen, Rui Chang, Jifeng Guo, Jerzy Baranowski, "Emotion Recognition of EEG Signals Based on the Ensemble Learning Method: AdaBoost", Mathematical Problems in Engineering, vol. 2021, pp. 1, 2021
3. G. Chanel, C. Rebetez, M. Bétrancourt, and T. Pun, “Emotion assessment from physiological signals for adaptation of game difficulty,” IEEE Transactions on Systems, Man, and Cybernetics A Systems and Humans, vol. 41, no. 6, pp. 1052–1063, 2011.
Eric Andrews Thank you for your informative response. Initially, we are attempting to reproduce an emotion classifier, moreso replicating existing models. Eventually, we intend to use the emotion classifier to generalize to new participants with a generative music algorithm.
Hence, the emotion classifier may not need to be trained on new participants if the bootstrapping approach you mentioned works. However, eventually, we intend to recruit participants that can feed through the model, and then be sent to a generative algorithm. A good summary of this attached (Ehrlich, 2019):
The only difference would be our design does not use an LDA model for emotion classification, as I believe Ehrlich's design does. We also need access to DEAP pre-processed data. We only have it within one of our classifier datasets. Do you have any idea how the preprocessed MATLAB data can be accessed?
- Badges
- Science topic
- Similar topics
- Correction