Motor-Imagery EEG


Zine Research Lab, NIT Jaipur


EEG or Electroencephalogram refers to recordings of electrical activity in the brain. Motor-Imagery EEG is generated when users simulate an action mentally, such as imagining your left hand move. Our objective was to recognize these signals from processed EEG using stochastic search algorithms, where we focussed on Genetic and Evolutionary algorithms.


This was one of my first research initiatives during 2nd year of engineering. With the guidance of seniors, I learned and implemented various such algorithms. After applying several published methods and additional improvements to some classic algorithms in this area(PSO,DEs,etc) we were able to formulate two promising algorithms which showed impressive results.

  1. BSA-NN : A randomized backtracking swarm on a three layer neural-net, with minimal parameter tuning and parallelized OVA(one vs. all multiclass implementation).

  2. GSEA : Iterative group based evolution and mutation scheme, dividing the swarm into multiple fitness groups.

GSEA Architecture


Accuracy of 69% across 3 different subjects on a multiclass motor-imagery problem. This performance is better than 21 previous approaches including the winning approach of BCI competition 3(dataset 2).


1. SK Agarwal, Saatvik Shah, and Rajesh Kumar. “Classification of mental tasks from EEG data using backtracking search optimization based neural classifier.” Neurocomputing (2015). URL

2. SK Agarwal , Saatvik Shah, and Rajesh Kumar. “Group based Swarm evolution algorithm (GSEA) driven mental task classifier.” Memetic Computing 7.1 (2015): 19-27. URL