Modeling and analysis tools for brain study

Abstract

Stress has become an inevitable element in our daily lives. An acceptable level of stress may assist human in one way or another, but excessive stress is devastating to health. Many methods can be used to monitor stress. The objective of this project is to develop integrated tools for modeling and analysis of stress. In this project, an algorithm for stress level recognition using Electroencephalogram (EEG) is suggested. Raw data was collected from 9 test subjects. 4 different levels of stress were induced into the test subjects using a Stroop color-word test and EEG raw data were recorded. Feature extraction methods, fractal dimension (FD), statistical features (Stats) and traditional power features (Power) were analyzed with different combinations. Then Multilayer Perceptron (MLP) was used as the classifier. 2 to 4 levels of stress can be recognized with different degree of accuracies. 4 levels of stress were recognized with an accuracy of 64.4% using FD and Stats, 3 levels 69.3% using all three feature extraction methods and 2 levels 83.0% using FD and Stats. The accuracy was improved after fine tuning MLP hyper parameters. The algorithm is later integrated into the system CogniMeter for stress level monitor. User’s degree of stress is reflected in real time basis. The system can be utilized in sectors such as air-traffic controllers, operators, etc. for stress monitoring.