Hello,

The question you have asked is sort of open ended. There is no definitive answer describing the best steps for pre-processing EDA signals. The quality of EDA signal varies depending on the location the signals are captured. Most common locations are finger tips and feet. Also, the type of sensor used to capture the signal contributes to its quality. For example contact based sensors are expected to produce better quality signal than contact less sensors (wrist worn sensors). Depending on all those factors, one should decide the ideal steps for pre processing the signal. That is the reason why you are confronted with different approach in every research. Well, none of those approaches are wrong, the question is which one would be most suited to your data set and your research objective. Hence, it would be good idea to consider the experimental settings and conditions on which the signals were collected.

However, In my research, I use EDA and PPG signals extensively. My EDA signals are collected from finger tips and by using contact based sensor. As you said that you have been researching about various approaches for pre processing, I won't ramble about the obvious facts about EDA signal. I will instead provide a brief overview regarding my approach for preprocessing the signal.

My research objective is identifying stress levels with the help of EDA signals which requires extraction of features like occurrence of peaks and their characteristics. Before feature extraction, I typically follow these steps:

1) Removing linear trends: When you are dealing with EDA signals of multiple subjects and using those signals for analysis, it is important to remove the linear trends associated with the signals. Actually this is common for any time series signals. But yes, I would say this would be the first step.

2) Isolating skin conductance level from skin conductance response: Usually, I would perform this step especially because it is important for my research problem, but as you said you only need the skin conductance level and not the response, I am not quite sure as why would you want that. It would be better if more information was provided regarding your research objective.

3) Filtering: Ideally I use low pass butterworth filter. The cut off frequency is usually determined by observing the power spectral density of the signal. Typically for EDA signal, the cut off frequency should be less than 4 Hz.

Usually these three steps should be sufficient for getting a desired quality signal. In summary I would say, your preprocessing is only as good as the processed signal you obtain. I would encourage you to keep exploring your signals and have an understanding of how your signal behaves and how well is it aligning with your research objective. The more you visualize, the better you understand, the better you will perform.

Rajdeep Kumar Nath Thank you very much for your time and your answer. I'm going to work with it. It's good to know why every approach is different and which approach you use for your research.

I want to know to what extent the empathy feelings from the respondents are measurable by bio-signals, specifically EDA signals.

Good luck with your research!