I'm working on a little project to measure Emotions with different Methods (K-NN, DNN, NN, Random Forest ...)
At the moment I have two data sets, with one of them I train the different methods (public DEAP data set) and then I want to test the results on the other data set.
Now I am trying to get the two data sets to a suitable level and I noticed that the second data set, which was recorded with the EMOTIV EPOC+, has large peaks in different places.
Is it possible that the peaks were caused by the blinking ?
Furthermore, both datasets are recorded with 128Hz and have as unit µV but the EPOC+ dataset has a value range of [-1000 ; 1000] and the DEAP dataset only a value range of [-10; 10], what could be the cause ?
I uploaded a plot from an example of each dataset.
Hi Lukas Kleybolte ,
The large peaks in the EPOC dataset do look like blinks/other eye movements. If you used the experiment outlined in the DEAP paper, you likely have a good deal of ocular artifacts in your data. For example, the large spike around ~24 seconds in the EPOC dataset looks like an eyeblink, whereas the large perterbations later on (38-47 seconds) look like generalized eye movement (participant scanning the image, moving eyes freely). Given the size of the artifacts in the EPOC dataset, it's not necessarily surprising that your y-axis is so large compared to the DEAP data. Of note, the EPOC data also have some pretty noticeable electrical artifact (the very fast oscillations seen across all channels from ~25-39 seconds).
I think with some basic artifact rejection and filtering you could knock out a lot of your artifacts. There are a number of approaches you could use depending on the additional sensors you used to record your data (e.g., EOG, earlobe/mastoid references, etc.). It looks like you are using the preprocessed DEAP data, which have already been corrected for these artifacts. To best match your data sets, I would recommend following the same procedures as outlined on the DEAP data website ( https://www.eecs.qmul.ac.uk/mmv/datasets/deap/readme.html ). Good luck!
Hi Lukas Kleybolte ,
The large peaks in the EPOC dataset do look like blinks/other eye movements. If you used the experiment outlined in the DEAP paper, you likely have a good deal of ocular artifacts in your data. For example, the large spike around ~24 seconds in the EPOC dataset looks like an eyeblink, whereas the large perterbations later on (38-47 seconds) look like generalized eye movement (participant scanning the image, moving eyes freely). Given the size of the artifacts in the EPOC dataset, it's not necessarily surprising that your y-axis is so large compared to the DEAP data. Of note, the EPOC data also have some pretty noticeable electrical artifact (the very fast oscillations seen across all channels from ~25-39 seconds).
I think with some basic artifact rejection and filtering you could knock out a lot of your artifacts. There are a number of approaches you could use depending on the additional sensors you used to record your data (e.g., EOG, earlobe/mastoid references, etc.). It looks like you are using the preprocessed DEAP data, which have already been corrected for these artifacts. To best match your data sets, I would recommend following the same procedures as outlined on the DEAP data website ( https://www.eecs.qmul.ac.uk/mmv/datasets/deap/readme.html ). Good luck!
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing