Multi-target way of cursor movement in brain computer interface using unsupervised learning

Brain-Computer Interfacing(BCI) helps physically disabled people to control multidimensional cursor movement in a real-life scenario. Noninvasive BCI techniques play a major role for this purpose. In this paper, we have proposed three algorithms of cursor movement using three well-known clustering methods (Minimum distance, DB-Scan and Gaussian Mixture Model). These proposed techniques are generally tested on motor imagery EEG data. We have performed multi-target based cursor movement using our previously proposed “FindTarget” algorithm (used for single target based cursor movement). However, we have also performed the comparative analysis of these three proposed algorithms based on internal & external validation indices and identified the best one for multi-target based cursor movement. The average accuracy of our proposed model is 68.4% (Kaggle dataset) and 70.36% (Openvibe dataset). We have evaluated the execution time of the proposed algorithms and found the most efficient proposed algorithm(O(n)) with respect to time and space for both the datasets. The result indicated that our proposed method gives more accuracy than all the previous methods of cursor movement. Our proposed method is more reliable, timely-efficient and experimental setup independent compared to other existing methods. If we choose the suitable algorithm based on time and space complexity then response time of multi-target based BCI system will be improved so that disabled people can effectively communicate with the external world.