EEG based cognitive task classification using multifractal detrended fluctuation analysis

Locating cognitive task states by measuring changes in electrocortical activity due to various attentional and sensory-motor changes, has been in research interest since last few decades. In this paper, different cognitive states while performing various attentional and visuo-motor coordination tasks, are classified using electroencephalogram (EEG) signal. A non-linear time-series method, multifractal detrended fluctuation analysis (MFDFA) , is applied on respective EEG signal for features. Using MFDFA based features a multinomial classification is achieved. Nine channel EEG signal was recorded for 38 young volunteers (age: $$ years, 30 male and 8 female), during six consecutive tasks. First three tasks are related to increasing levels of selective focus vision; next three are reflex and response based computer tasks. Total of 90 features (ten features from each of nine channel) were extracted from Hurst and singularity exponents of MFDFA on EEG signals. After feature selection, a multinomial classifier of six classes using two methods: support vector machine (SVM) and decision tree classifier (DTC). An accuracy of 96.84% using SVM and 92.49% using DTC was achieved.