Abstract:
The EEG signals captured from central nervous system, have the property to directly
reflect the brain activity and thus have an intrinsic relationship with the emotional states
of humans. Researchers are paying more attention to emotion recognition from EEG sig nals with the development of Brain Computer Interface (BCI) technology. BCI is a direct
communication or mapping between the electrical activity of the brain and external devices.
Emotion recognition gained more attention in the past decade as it is directly connected
to various fields like psychology, healing, physiology, marketing and studies. The gaming
market has expanded to be one of the most significant entertainment markets because of
recent advanced technologies. Understanding the automatically inferred player’s emotions
during game play can be used to enhance the quality of the game. This work aims to iden tify the emotions using a reliable LSTM network from the selected features of EEG with a
better recognition rate. Interactive software like video games can elicit a range of feelings in
their users. Here in this work, GAMEEMO dataset is used where the EEGs are in response
to 4 different game plays. A video game can affect players’ thoughts and feelings through
game play, storytelling, and the gaming environment. At first the extraction of maximum
features using a suitable feature extraction method from the dataset is done. Then emotion
recognition using different machine learning models and recurrent neural networks are done
followed by their performance comparison. The LSTM model emerged as the one with bet ter performance . The model is then used for emotion detection from selected features of
the dataset to study the effect of feature selection on the performance. The results shows
performance improvement of the model with selected features. Finally, the model was used
on various experimental datasets like EEG Brainwave Dataset and an experimental sample
dataset from DEAP to study its applicability and reliability
