Using Probabilistic Direct Multi-class Support Vector Machines to Improve Mental States Based-Brain Computer Interface

摘要: Brain-Computer Interface (BCI) system allows physically challenged people to operate with their external surroundings, just through brain signals. Since the objective of BCI is categorize signals into homogeneous classes each which represents a mental state, it necessary choose an appropriate discrimination approach. So, we use Support Vector Machines (SVM) due multiple benefits. The SVM are suggested treat binary problems, conversion multiclass cases (M-SVM) includes: indirect methods based on decomposition approaches, and direct that consider all simultaneously. This experiment aims introduce four existing M-SVM in problematic states recognition. discriminators independently give probability estimates relative five states. Results indicate models generate nearly similar accuracies. Nevertheless, average rates ranging from 68.25% 90.86%, Crammer Singer discriminator outperforms other models.