Evolutionary Weighted Ensemble for EEG Signal Recognition

Recognition of an EEG signal is a very complex but very important problem. In this paper we focus on a simplified classification problem which consists of detection finger movement based on an analysis of seven EEG sensors. The signals gathered by each sensor are subsequently classified by the respective classification algorithm, which is based on data compression and so called LZ-Complexity. To improve overall accuracy of the system, the Evolutionary Weighted Ensemble (EWE) system is proposed. The parameters of the EWE are set in a learning procedure which uses an evolutionary algorithm tailored for that purpose. To take full advantage of information returned by sensor classifiers, setting negative weights are permitted, which significantly raises overall accuracy. Evaluation of EWE and its comparison against selected traditional ensemble algorithm is carried out using empirical data consisting of almost 5 hundred samples. The results show that the EWE algorithm exploits the knowledge represented by the sensor classifiers very effectively, and greatly improves classification accuracy.

[1]  Bonggun Shin,et al.  Bayesian Group Nonnegative Matrix Factorization for EEG Analysis , 2012, ArXiv.

[2]  Yoav Freund,et al.  Boosting a weak learning algorithm by majority , 1995, COLT '90.

[3]  Michal Prilepok,et al.  Electroencephalogram classification methods , 2013 .

[4]  Michal Wozniak,et al.  Some Remarks on Chosen Methods of Classifier Fusion Based on Weighted Voting , 2009, HAIS.

[5]  Álvaro Herrero,et al.  International Joint Conference CISIS'12-ICEUTE'12-SOCO'12 Special Sessions, Ostrava, Czech Republic, September 5th-7th, 2012 , 2013, CISIS/ICEUTE/SOCO Special Sessions.

[6]  Zbigniew Michalewicz,et al.  Handbook of Evolutionary Computation , 1997 .

[7]  C. K. Chow,et al.  Statistical Independence and Threshold Functions , 1965, IEEE Trans. Electron. Comput..

[8]  Sherif Hashem,et al.  Optimal Linear Combinations of Neural Networks , 1997, Neural Networks.

[9]  David G. Stork,et al.  Pattern Classification , 1973 .

[10]  Yoav Freund,et al.  Experiments with a New Boosting Algorithm , 1996, ICML.

[11]  Kazuyuki Murase,et al.  Ensembles of Neural Networks Based on the Alteration of Input Feature Values , 2012, Int. J. Neural Syst..

[12]  Václav Snásel,et al.  Compression-based similarity in EEG signals , 2013, 2013 13th International Conference on Intellient Systems Design and Applications.

[13]  Anil K. Jain,et al.  Statistical Pattern Recognition: A Review , 2000, IEEE Trans. Pattern Anal. Mach. Intell..

[14]  Ethem Alpaydin,et al.  Introduction to machine learning , 2004, Adaptive computation and machine learning.

[15]  Seungjin Choi,et al.  Semi-Supervised Nonnegative Matrix Factorization , 2010, IEEE Signal Processing Letters.

[16]  Bogdan Gabrys,et al.  Classifier selection for majority voting , 2005, Inf. Fusion.

[17]  Bogdan Gabrys,et al.  Genetic algorithms in classifier fusion , 2006, Appl. Soft Comput..

[18]  Ludmila I. Kuncheva,et al.  Combining Pattern Classifiers: Methods and Algorithms , 2004 .

[19]  David G. Stork,et al.  Pattern Classification (2nd ed.) , 1999 .

[20]  Leo Breiman,et al.  Bagging Predictors , 1996, Machine Learning.

[21]  Marek Penhaker,et al.  Pattern Recognition in EEG Cognitive Signals Accelerated by GPU , 2012, CISIS/ICEUTE/SOCO Special Sessions.