Electrodermal activity, also known as skin conductance, is a psychophysiological signal measured in human subjects in studies aimed at better understanding fear, anxiety, stress and decision- making. This measurement is helpful for cognitive-science researchers investigating all facets of human behaviour. However, they don’t know precisely how to extract the information from the signal to draw valid conclusions.

Imagine you’re browsing through a cookbook by one of the world’s finest pastry chefs. Your eyes come to rest on a recipe for chocolate cake that is so appetizing your stomach starts growling. You decide to try your hand at preparing this tantalizing treat. Unfortunately, though, there’s some specific information missing from the recipe, and you can’t see how to make the dessert properly. You follow the recipe anyway, and the resulting cake isn’t at all the way it looks in the picture. Since you know the outcome, and this isn’t your first attempt at baking a cake, you realize that you could have improvised or even tweaked the recipe at an opportune time.

Now replace the pastry chef by a scientist, the recipe by a scientific paper, and the cake by the conclusions of the paper. You read that right: conducting a science experiment can be compared to baking a cake by following the steps of a recipe. One crucial difference, though, is that in science, improvising or tweaking the “recipe” based on experience and knowledge—especially if the goal is to replicate the results of a particular study—is a riskier proposition. If a research team does want to replicate a prior study, they must follow exactly the same methodology as in that original study. Otherwise, they cannot be sure that their findings point to a valid phenomenon and are not simply the product of chance.

I use the recipe analogy because the aim of my research project is to establish a clear, precise method for the pre-processing of the electrodermal activity (EDA) signal. More specifically, the EDA signal measures variations in skin conductance using a pair of electrodes placed on the skin. Skin has very high electrical resistance; i.e., it does not conduct electricity easily. Fortunately, sweat released through the skin increases its conductance. Sweat is released by glands under the skin, which are controlled by specific areas of the brain that activate when a person is stressed, anxious, afraid, or trying to make a decision, among many other situations. The EDA signal is used by psychology and cognitive neuroscience researchers investigating certain types of human behaviour, including those mentioned above.

There are, however, still shortcomings in research when it comes to EDA signal pre-processing. Like other research teams, I am trying to remedy those shortcomings. To that end, my goal is to develop a reliable method. The 2020 paper Filtering and model-based analysis independently improve skin-conductance response measures in the fMRI environment: Validation in a sample of women with PTSD, by Anthony A. Privratsky of the University of Arkansas for Medical Sciences and his colleagues, provides a good illustration of the shortcomings involved in filtering of the EDA signal by researchers.

In that paper, Privratsky et al. established a method, or “recipe,” for pre-processing of the EDA signal, which serves as a guideline. My project is based on that method. It has three objectives: first, add a step to their “recipe” and determine whether that improves the results. To achieve this, I am using artificial intelligence, specifically a machine-learning algorithm that detects unusable portions of the signal resulting from so-called artifacts. This may seem strange, but when there are variations in contact between the electrode and the skin, this creates what is known as a movement artifact. The problem with artifacts is that they are similar to the valid signal—the one the researchers are interested in. Studying these artifacts is not very useful because they do not result from the stimuli that are of the experiment, but from the subject’s movements.

Privratsky et al. use two ways of assigning a score to the EDA signal. These scores allow researchers to conduct statistical analyses to answer the questions posed in their study. You can see these scores as an objective means of judging a cake: did it taste good or not? In the case of EDA, the scores allow evaluation of the subject’s reaction to the stimuli. My second objective is to use another method of grading the signal and assess whether that improves the results. My third objective is to test Privratsky et al.’s “recipe” using the same ingredients, but sourced from different suppliers. In other words, I am testing the method, but using different publicly accessible databases.

My goal is therefore to produce a “recipe” enabling proper pre-processing of the EDA signal, but also to provide tools along with the recipe. In other words, I hope to provide future research teams with the tools I used and/or developed to complete the project; i.e., all of the code generated, but also the proper method of using it.