Figure 1 shows our pipeline architecture for preparing the data and extracting the set of features for classification. There are 3 different sections in this pipeline: (1) Data Preparation, (2) pyEDA [28], (3) Post Feature Extraction.

Galvanic skin response (GSR) feature extraction pipeline. EDA: electrodermal activity.

The primary purpose of the Data Preparation in our pipeline is to synchronize the data with the labels. To prepare the data for feature extraction, we extracted the original signals’ slices that match with their corresponding labels. With this aim, the slices of GSR data and their labels are collected in this part to be fed to the pyEDA for pre-processing and feature extraction.

The architecture of the pyEDA is shown in Figure 2. According to this figure, Preprocessing and Feature Extraction are the 2 main stages in this pipeline.

Pipeline architecture of the pyEDA. EDA: electrodermal activity; GSR: galvanic skin response.

In the preprocessing stage of the pyEDA pipeline, at first, the data are down-sampled; then, a moving average is used to smooth the data and reduce the artifacts such as body gestures and movements. In the end, the data are normalized to become suitable for classification models.

If the GSR data are collected at 128 Hz, it can safely be down-sampled to a 20 Hz sampling rate. This down-sampling has been done to conserve memory and processing time of the data. In this work, we did not down-sample the data since the original data are already sampled at 4 Hz, which is good in terms of time and memory usage.

In this work, several steps were taken to remove motion artifacts from the GSR signal. First, we used a moving average across a 1-second window to remove the motion artifacts and smooth the data. Second, a low-pass Butterworth filter on the phasic data was applied to remove the line noise. Lastly, preprocessed GSR signals corresponding to each different pain level were visualized to ensure the validity of the signals.

The pyEDA uses 2 different algorithms for feature extraction (Statistical Feature Extraction and Deep Learning Feature Extraction). The parameters of the Deep Learning Feature Extraction part of the pipeline are set and tuned for stress detection; therefore, in this work, we only used the features extracted by the Statistical Feature Extraction algorithm.

The number of peaks, the mean, and the max peak amplitude are the 3 different statistical features that are extracted in the pyEDA. The GSR signals consist of 2 main components: skin conductance level, also known as the tonic level of GSR, and skin conductance response, also called the phasic component of GSR. The GSR peaks or bursts are considered the variations in the phasic component of the signal. Therefore, the most important part in extracting the peaks of the GSR signal is to extract its phasic component. Based on Figure 2, the pyEDA tool uses the cvxEDA algorithm [29] to extract the phasic component. Then, the phasic component and the preprocessed GSR data are fed to the Statistical Feature Extraction module to extract the 3 mentioned features (number of peaks, mean GSR, and max peak amplitude).

We also extracted the features that were used in the work by Werner et al [13] for the GSR signals. The preprocessed GSR signals and the set of features (number of peaks, mean GSR, and max peak amplitude) were fed into the Post Feature Extraction module to extract these features.

The maximum value of the peaks, range, standard deviation, interquartile range, root mean square, mean value of local maxima, mean value of local minima, mean of the absolute values of the first differences, and mean of the absolute values of the second differences are the extra features that were extracted in this part. Table 1 shows all the extracted features with their descriptions.

Extracted galvanic skin response (GSR) features with their descriptions.

The mean of the absolute values of the first differences (mavfd) is calculated as:

The mean of the absolute values of the second differences (mavsd) is calculated as:

Note: The content above has been extracted from a research article, so it may not display correctly.



Q&A
Please log in to submit your questions online.
Your question will be posted on the Bio-101 website. We will send your questions to the authors of this protocol and Bio-protocol community members who are experienced with this method. you will be informed using the email address associated with your Bio-protocol account.



We use cookies on this site to enhance your user experience. By using our website, you are agreeing to allow the storage of cookies on your computer.