A Comparison Of Incremental Learning In Electroencephalography (EEG) Signal For Person Authentication Modelling

Soo, Pheng Kian (2015) A Comparison Of Incremental Learning In Electroencephalography (EEG) Signal For Person Authentication Modelling. Project Report. UTeM, Melaka, Malaysia. (Submitted)

Text (24 Pages) A Comparison Of Incremental Learning In Electroencephalography (EEG) Signal For Person Authentication Modelling 24 Pages.pdf - Submitted Version Download (292kB)

Abstract

EEG applications commonly use small training data for analysis due to limited recording time. Besides, the consistency EEG signals of an individual can be affected by environmental factor or attention shift. Thus, incremental model is suitable for EEG analysis due to its capability of adaptation. However, there exists little research work focusing in this area especially on person authentication modelling. This project aims to compare the performance of the proposed Incremental Support Vector Machine, Incremental K-Nearest Neighbour and Hoeffding Tree for person authentication modelling. The experimental data involves VEP signals from 10 common human subjects recorded from using 10-20 system. Electrodes PO7, PO3, POZ, PO4, PO8, O1, OZ, O2 were used for recording EEG dataset. Feature extraction i.e. mean, coherence, cross-correlation, mutual information, wavelet packet decomposition (WPD) and hjorth parameter has been done on the recorded EEG dataset. The data were divided to 20 percent for training set while 80 percent for testing set. WEKA Knowledge Work Flow was used for incremental classification task for Incremental K-Nearest neighbour and Hoeffding Tree while Incremental Support Vector Machine was implemented in Matlab environment. The measurement of accuracy and true positive detection rate were used as the performance measure among for the comparison methods. Statistical tests i.e. the Shapiro-Wilk Normality test, Friedman Test and ANOVA test were used for validation purposes. From the statistical test and result analysis, Incremental Support Vector Machine showed the best performance among other models. This is because Incremental Support Vector Machine can handle EEG dataset with multi-class, polarity and many feature data. Incremental K-Nearest Neighbour and Hoeffding Tree proven equally good in the validation test. Nevertheless, hybrid Incremental Support Vector Machine model with Hoeffding Tree Model is suggested in the future work to overcome the shortcoming of Incremental Support Vector Machine in handling unbalanced class in person authentication framework.

Actions (login required)

View Item