GA-SVM based feature selection and parameters optimization for BCI research

GA-SVM based feature selection and parameters optimization for BCI research,10.1109/ICNC.2011.6022083,Lei Wang,Guizhi Xu,Jiang Wang,Shuo Yang,Lei Guo,

GA-SVM based feature selection and parameters optimization for BCI research  
BibTex | RIS | RefWorks Download
Brain Computer Interface (BCI) can translate the mind of the patients who suffered from locked- in syndrome into control commands or meaning symbols. Using this technology, the patients can communicate with the world. The core parts of a typical BCI system is feature extraction and pattern recognition. Too many irrelevant and redundant features will increase the time of classification and decrease the prediction accuracy. The kernel parameters setting for support vector machine (SVM) also impact on the classification accuracy. In this paper, after the features extracted though the algorithm called Sample Entropy, GA-SVM hybrid algorithm was used with two purposes: Selecting of the optimal feature subset and deciding the parameters for SVM classifier. Compared with GA-based feature selection and GA-based parameters optimization for SVM, the GA-SVM hybrid algorithm has fewer input features and gain much higher classification accuracy. During the BCI research, feature extraction and pattern recognition play important roles. Selecting the useful features and getting rid of non-related features, optimizing the parameters of classifier will increase the classification speed and enhance the classification accuracy. Accord the lectures based on GA method used in feature selection and parameter optimization in other research areas, GA as a feature selection and optimization method be investigated in this paper(4~5). Compared with the result use GA only in feature selection and use GA only in classifier parameter optimization, the GA-SVM hybrid algorithm using fewer features, got a higher accuracy and showed great advantages.
Published in 2011.
Cumulative Annual
View Publication
The following links allow you to view full publications. These links are maintained by other sources not affiliated with Microsoft Academic Search.