EEG Signal Classification Using Fuzzy-Rough Nearest Neighbours (FRNN) Model For Person Authentication

Liew, Siaw Hong (2013) EEG Signal Classification Using Fuzzy-Rough Nearest Neighbours (FRNN) Model For Person Authentication. Project Report. UTeM, Melaka, Malaysia. (Submitted)

	PDF (24 Pages) EEG_Signal_Classification_Using_Fuzzy-Rough_Nearest_Neighbours_(FRNN)_Model_For_Person_Authentication.pdf - Submitted Version Download (327kB)
	PDF (Full Text) EEG_Signal_Classification_Using_Fuzzy-Rough_Nearest_Neighbours_(FRNN)_Model_For_Person_Authentication.pdf - Submitted Version Restricted to Registered users only Download (1MB)

Abstract

EEG signals especially Visual Evoked Potential (VEP) is unique but highly uncertain and difficult to process. Thus, identify the appropriate feature vector and prediction model are essential to implement this modality for person authentication purposes. The research on analyzing VEP for person authentication using soft computing modeling is very limited and hardly emphases on uncertainty methods even though uncertainty modeling has been proven efficient in many other domains. Fuzzy-Rough Nearest Neighbours (FRNN) model is outstanding to model uncertainty in element belongings under an imperfect data condition. This advantage is vital for person authentication modeling using VEP, but there is a lack of research work focusing in this direction. Hence, the objectives of this project are to identify the VEP active electrodes and significant feature vectors for authentication modeling, and to evaluate the performance of the proposed Fuzzy-Rough Nearest Neighbours (FRNN) model for person authentication classification. This project followed the experimental methodology including the preliminary studies, data preparation, EEG signals feature extraction, experimentation and result analysis. Mean, cross-correlation and coherence are the feature vectors that extracted from the lateral and midline electrodes. The classification results of FRNN using implicator and t-norm were promising against its comparison techniques of D-kNN and FLR especially in the measurement of AUC. Nevertheless, feature selection is suggested in the future work to minimize the dimension of data in order to achieve a better generalized feature space in the authentication framework.

Actions (login required)

View Item