Kohonen Neural Network Stress Detection Using Only Electrodermal Activity Features

Abstract

This paper presents a method for identifying human stress levels by using a Kohonen neural network. The study focuses on differentiating between a relaxed and a stressed state and it presents a series of parameters (skin conductance response signal power, skin conductance response signal frequency, skin conductance level gradient, response rise time and response amplitude) extracted only from the electrodermal activity signal. A very strict recording protocol was used to minimize the artifacts caused by the bad connection between electrodes and skin. A stress inducing method is presented that can be used to replicate results in laboratory conditions.

Keywords

• biomedical signal processing
• data analysis
• electrophysiology
• pattern recognition
• self organizing feature maps