Study on Accuracy Improvement of Mental Arithmetic Task Classification Using Different Classifiers with DWT Feature Extraction Method

Abstract

Near-infrared spectroscopy (NIRS) is a recently developed technique that can reveal hemodynamic and metabolic changes during cortical activation. NIRS has been used during cognitive tasks to study hemodynamic responses such as the change of oxyhemoglobin concentration. In the field of Brain Computer Interfacing (BCI), the use of fNIRS is an efficient approach. In this paper, fNIRS data from mental arithmetic tasks were proposed to classify with the help of the Discrete Wavelet Transform (DWT) based feature extraction method along with different classifiers. Raw data was preprocessed at first and stored in different frames to analyze brain activity. Using both the approximate and detail coefficients of DWT for framed data, features were extracted and used to compare brain activity during the mental arithmetic tasks and rest conditions. Finally, efficiencies of oxyhemoglobin, deoxyhemoglobin, and total hemoglobin data were measured for different channel combinations, and a satisfactory level of 95.54 % accuracy was achieved with the GentleBoost algorithm for the HAAR wavelet.