PERFORMANCE EVALUATION OF CLASSIFIERS IN DISTINGUISHING MENTAL TASKS FROM EEG SIGNALS

Isaak Kavasidis, Carmelo Pino, Concetto Spampinato, Francesco Maiorana

References

  1. [1] D. Krusienski, M. Grosse-Wentrup, F. Galan,D. Coyle, K. Miller, E. Forney, and C. An-derson, “Critical issues in state-of-the-art brain-computer interface signal processing,” Journalof Neural Engineering, vol. 8, no. 2, p. 025002,2011. [Online]. Available: http://eprints.pascal-network.org/archive/00007839/
  2. [2] L. Zhiwei and S. Minfen, Classification of Men-tal Task EEG Signals Using Wavelet Packet En-tropy and SVM, 2007, pp. 3906–3909.
  3. [3] N. J. H. Y. P. A. Yong and G. C. M. Sil-vestre, “Single-trial eeg classification for brain-computer interface using wavelet decomposi-tion,” 2005.
  4. [4] L. Guo, D. Rivero, J. A. Seoane, and A. Pazos,“Classification of eeg signals using relativewavelet energy and artificial neural networks,”175k-fold Cross Validation Error ratesClassifier 3 5 10 20 30 LO Without Cross ValidationLoglc 0.406 0.463 0.417 0.405 0.406 0.394 0.123Fisherc 0.428 0.394 0.388 0.406 0.393 0.417 0.116Nmc 0.701 0.690 0.759 0.731 0.731 0.842 0.225Perlc 0.463 0.455 0.509 0.491 0.538 0.537 0.101Lmnc 0.352 0.447 0.449 0.397 0.381 0.398 0.101Rbnc 0.513 0.494 0.489 0.557 0.488 0.518 0.172Knnc 0.647 0.652 0.703 0.647 0.710 0.774 0.232Parzenc 0.739 0.781 0.764 0.748 0.754 0.743 0.219Treec 0.442 0.456 0.438 0.445 0.421 0.473 0.173Quadrc 0.545 0.505 0.493 0.504 0.487 0.487 0.219Naivebc 0.508 0.496 0.524 0.520 0.543 0.520 0.154Kernel 0.588 0.635 0.647 0.578 0.600 0.583 0.199Parzendc 0.433 0.399 0.410 0.404 0.416 0.405 0.167Klldc 0.456 0.398 0.388 0.399 0.417 0.405 0.135Pcldc 0.444 0.393 0.428 0.398 0.387 0.405 0.135Polyc 0.387 0.388 0.399 0.411 0.399 0.417 0.135Subsc 0.590 0.659 0.584 0.619 0.636 0.624 0.242Table 2. Error with and without Cross ValidationFeatures Selection methodsClassifier Fisher Loglc Klldc Perlc Without features selectionLoglc 0.104 0.078 0.097 0.111 0.123Fisherc 0.091 0.137 0.123 0.137 0.116Nmc 0.124 0.205 0.201 0.155 0.225Perlc 0.196 0.196 0.129 0.189 0.101Lmnc 0.116 0.148 0.102 0.134 0.101Rbnc 0.135 0.162 0.155 0.123 0.172Knnc 0.084 0.254 0.192 0.161 0.232Parzenc 0.103 0.241 0.223 0.236 0.219Treec 0.110 0.180 0.187 0.198 0.173Quadrc 0.172 0.136 0.142 0.223 0.219Naivebc 0.142 0.155 0.155 0.156 0.154Kernel 0.109 0.174 0.160 0.180 0.199Parzendc 0.128 0.204 0.135 0.222 0.167Klldc 0.103 0.123 0.084 0.111 0.135Pcldc 0.103 0.123 0.110 0.111 0.135Polyc 0.104 0.150 0.214 0.143 0.135Subsc 0.161 0.176 0.152 0.137 0.242Table 3. Classification error rates with and without features selection176Features Selection methodsCombiner Fisher loglc klldc perlc Without features selectionProdc 0.180 0.294 0.301 0.281 0.282Meanc 0.129 0.179 0.123 0.160 0.135Median 0.084 0.156 0.129 0.205 0.173Maxc 0.129 0.149 0.123 0.161 0.141Minc 0.098 0.205 0.201 0.155 0.225Votec 0.078 0.117 0.097 0.143 0.173Parallel 0.129 0.149 0.123 0.161 0.141Stacked 0.129 0.149 0.123 0.161 0.141Table 4. Error rates when combiners are used with and without feature selectionProceedings of the first ACMSIGEVO Sum-mit on Genetic and Evolutionary ComputationGEC 09, p. 177, 2009. [Online]. Avail-able: http://portal.acm.org/citation.cfm?doid=1543834.1543860
  5. [5] C.-J. Lin and M.-H. Hsieh, “Classification ofmental task from eeg data using neural networksbased on particle swarm optimization,” Neuro-computing, vol. 72, no. 4-6, pp. 1121–1130,2009.
  6. [6] N.-Y. Liang, P. Saratchandran, G.-B.Huang, and N. Sundararajan, “Classificationof mental tasks from eeg signals usingextreme learning machine.” InternationalJournal of Neural Systems, vol. 16,no. 1, pp. 29–38, 2006. [Online]. Available:http://www.ncbi.nlm.nih.gov/pubmed/16496436
  7. [7] E. Forney and C. Anderson, “Classification ofeeg during imagined mental tasks by forecastingwith elman recurrent neural networks,” in Neu-ral Networks (IJCNN), The 2011 InternationalJoint Conference on, 31 2011-aug. 5 2011, pp.2749 –2755.
  8. [8] V. Khare, J. Santhosh, S. Anand, andM. Bhatia, “Performance comparison ofthree artificial neural network methods forclassification of electroencephalograph signalsof five mental tasks,” Online, vol. 2010, no.February, pp. 200–205. [Online]. Available:http://www.scirp.org/Journal/PaperDownload.aspx?paperID=1337&fileName=JBiSE20100200011 58633660.pdf
  9. [9] A. Faro, D. Giordano, M. Pennisi, G. Scarcio-falo, C. Spampinato, and F. Tramontana, “Tran-scranial magnetic stimulation (tms) to evaluateand classify mental diseases using neural net-works,” in Artificial Intelligence in Medicine,ser. Lecture Notes in Computer Science, 2005,vol. 3581, pp. 310–314.
  10. [10] A. Crisafi, D. Giordano, and C. Spampinato,“Griplab 1.0: Grid image processing laboratoryfor distributed machine vision applications,” En-abling Technologies, IEEE International Work-shops on, vol. 0, pp. 188–191, 2008.
  11. [11] A. Faro, D. Giordano, and F. Maiorana, “Min-ing massive datasets by an unsupervised paral-lel clustering on a grid: Novel algorithms andcase study,” Future Generation Computer Sys-tems, vol. 27, no. 6, pp. 711–724, 2011.
  12. [12] D. Giordano, “Evolution of interactive graphi-cal representations into a design language: a dis-tributed cognition account,” International Jour-nal of Human-Computer Studies, vol. 57, no. 4,pp. 317–345, 2002.
  13. [13] A. Faro, D. Giordano, and C. Santorno, “Link-based shaping of hypermedia webs assisted bya neural agent,” Journal of Universal ComputerScience, vol. 4, no. 7, pp. 630–651, jul 1998.
  14. [14] A. Faro, D. Giordano, F. Maiorana, andC. Spampinato, “Discovering genes-diseases as-sociations from specialized literature using thegrid,” IEEE Trans Inf Technol Biomed, vol. 13,pp. 554–560, Jul 2009.
  15. [15] A. Faro, D. Giordano, and C. Spampinato,“Combining literature text mining with microar-ray data: advances for system biology model-ing,” Brief Bioinform, Jun 2011.177

Important Links:

Go Back