Psychophysiology Meets Computer Science: Predicting the Magnitude of Participant Physiological Response with Machine Learning

The present inquiry uses methods from psychophysiology and machine learning to reduce overall error in classification models. The field of psychophysiology, though rooted in decades of experimentation, has never reached the same level of precision as some aspects of medical inquiry. In fact, while some medical regression models, when determining some way to classify a patient\u27s illness based on certain symptoms, can result in highly significant results with large effect sizes, equal levels are virtually unheard of in psychophysiology. The present investigation attempts to unravel some part of this mystery and determines some possible reasons for the difficulty in finding similar effect sizes, especially concerning methods that match participant state with physiological response. Of particular focus are two areas: baseline research and experimental data analysis methods. The role of baselining techniques in relation to overall quality of response is the first emphasis and this interest stems from the Law of Initial Value that indicates some relationship between baseline and experimental response. Though this relationship has been continually investigated and found to be lacking for many physiological measures, experimental condition heart rate response has been consistently shown to rely heavily on baseline response. This finding influences the second half of the present inquiry, which deals with the overall analysis of experimental data and the role that traditional statistics could play in the present problem. By comparing logistic regression and support vector models, it is expected that researchers would use the preferred method, based on their goals, to flag potentially highly influential cases that may greatly skew data and make modeling difficult. Additionally, demographic characteristics that could also help identify these influential cases in the future before modeling are shown