PERSON IDENTIFICATION USING FUNCTIONAL NEAR- INFRARED SPECTROSCOPY SIGNALS USING A FULLY CONNECTED DEEP NEURAL NETWORK

Authors : Ozge MERCANOGLU SINCAN, Hacer YALIM KELES, Yagmur KIR, Adnan KUSMAN, Bora BASKAK

Pages : 55-68

View : 22 | Download : 6

Publication Date : 2017-12-21

Article Type : Research Paper

Abstract :In this study, we investigate the suitability of functional near-infrared spectroscopy signals insert ignore into journalissuearticles values(fNIRS); for person identification using data visualization and machine learning algorithms. We first applied two linear dimension reduction algorithms: Principle Component Analysis insert ignore into journalissuearticles values(PCA); and Singular Value Decomposition insert ignore into journalissuearticles values(SVD); in order to reduce the dimensionality of the fNIRS data. We then inspected the clustering of samples in a 2d space using a nonlinear projection algorithm. We observed with the SVD projection that the data integrity associated with each person is high in the reduced space. In the light of these observations, we implemented a random forest algorithm as a baseline model and a fully connected deep neural network insert ignore into journalissuearticles values(FCDNN); as the primary model to identify person from their brain signals. We obtained %85.16 accuracy with our FCDNN model using SVD reduction. Our results are in parallel with the neuroscience researches, which state that brain signals of each person are unique and can be used to identify a person.
Keywords : functional near infrared spectroscopy fNIRS, person identification, PCA, SVD, fully connected deep neural network, random forest