Date of Award

3-21-2013

Document Type

Thesis

Degree Name

Master of Science

Department

Department of Electrical and Computer Engineering

First Advisor

Jeffrey D. Clark, PhD.

Abstract

The ability to identify a stressed person is becoming an important aspect across different work environments. Especially in higher-stress career fields, such as first responders and air traffic controllers, mental stress can inhibit a person's ability to accomplish their job. A person's efficiency and psychological state in the work environment can be impeded due to poor mental health. Stress can result in harmful effects on the body, both physically and mentally, including depression, lack of sleep, and fatigue, which can lead to reduced work productivity. Research is being conducted to detect stress in workload-intensive environments. This thesis implements an imaging approach that utilizes hyperspectral data across the visible through shortwave infrared electromagnetic spectrum. The data is applied to the feature selection algorithms ReliefF, Support Vector Machine Attribute Evaluator (SVM AE), and Non-Correlated Aided Simulated Annealing Feature Selection-Integrated Distribution Function (NASAFS-IDF) to obtain features that discriminate between the classes, stress and non-stress. This data is classified using naive Bayes, Support Vector Machine (SVM), and decision tree methodologies. The feature set and classifier that produce the highest classification results are calculated using percent accuracy and area under the curve (AUC). The reported results are divided into contact and non-contact (NC) validation sets. The contact validation returned a high accuracy of 96.30% and high AUC of 0.979. Validation on NC models returned a high accuracy of 99.64% and high AUC of 0.998.

AFIT Designator

AFIT-ENG-13-M-38

DTIC Accession Number

ADA584003

Share

COinS