"Electrodermal Activity artifact correction BEnchmark" (EDABE) is a dataset for training and testing artifact recognition and correction models to automatically remove major artifacts in electrodermal activity (EDA) signals. It is the first public benchmark to compare methods. EDABE contains a total of 74.46 h of EDA recording affected by hand and body motion artifacts from 43 subjects. It is divided into a training set with 33 subjects (56.27 h), and test set with 10 subjects (18.19 h). The data was collected using a Shimmer3 GSR+ Unit at 128 Hz. The dataset is used to develop a fully automatic pipeline that emulates the manual correction done by the expert, providing a final clean signal. The paper that describe the pipeline is currently in a peer-review process. Each file includes in the filename the user_id and the expert that correct the signal. In addition, the file includes the signal with the following variables: time: timestamp of the signal. rawdata: raw data obtained by Shimmer3 GSR+ Unit. cleandata: reconstructed clean signal performed by a human expert. binarytarget: label of each sample as artefact or no artifact. signal_automatic: automatic cleaning of the signal performed by the automatic pipeline. predArtifacts: label predicted by the automatic cleaning pipeline. postProcessedPredArtifacts: label predicted by the automatic cleaning pipeline after postprocessing.

The research leading to this dataset has received partial funding from the European Commission (RHUMBO H2020-MSCA-ITN-2018-813234), and from the Universitat Politècnica de València (PAID-10-20)