A Data Driven Empirical Iterative Algorithm for GSR Signal Pre-Processing

In this paper, we introduce a data driven iterative low pass filtering technique, the Empirical Iterative Algorithm (EIA) for Galvanic Skin Response (GSR) signal preprocessing. This algorithm is inspired on Empirical Mode Decomposition (EMD), with performance enhancements provided by applying Midpoint-based Empirical Decomposition (MED), and removing the sifting process in order to make it computational inexpensive while maintaining effectiveness towards removal of high frequency artefacts. Based on GSR signals recorded at the wrist we present an algorithm benchmark, with results from EIA being compared with a smoothing technique based on moving average filter - commonly used to pre-process GSR signals. The comparison is established on data from 20 subjects, collected while performing 33 different randomized activities with right hand, left hand and both hands, respectively. In average, the proposed algorithm enhances the signal quality by 51%, while the traditional moving average filter reaches 16% enhancement. Also, it performs 136 times faster than the EMD in terms of average computational time. As a show case, using the GSR signal from one subject, we inspect the impact of applying our algorithm on GSR features with psychophysiological relevance. Comparison with no preprocessing and moving average filtering shows the ability of our algorithm to retain relevant low frequency information.