A comparative assessment of the simple and modified Randles cells using machine learning for optimal impedance fitting on artificial skin

Abstract Wearable devices in the healthcare space rely on electrodes interfacing with the skin to capture biosignals such as the electroencephalogram (EEG), electromyogram (EMG), and electrocardiogram (ECG), and a low electrode–skin impedance is essential for high-fidelity recording. Equivalent circuit models (ECMs) are widely used to interpret impedance data, but choosing models that balance accuracy with interpretability remains challenging. In this study, impedance spectra from seven types of commercial gelled and dry electrodes interfacing with artificial skin were fitted using two ECMs: the common Randles cell and a modified Randles cell featuring a Warburg impedance element. Both ECMs fit the impedance data well, but logistic regression (LR) using the ECM parameters revealed key differences. While either model could be used to classify electrodes as gelled or dry with > 70% accuracy, only the modified Randles cell supported multiclass differentiation among the seven electrode types with > 80% accuracy. These findings demonstrate that a Warburg-modified Randles cell results in more informative fitting results during wearable sensor characterization. More broadly, this report establishes a generalizable workflow integrating impedance fitting with supervised machine learning (ML) to optimize ECM selection, which we strongly recommend to researchers in the wearable devices field.