Applying Machine Learning to Predict Electron Transfer Kinetics from Voltammetry Experiments

Electrochemistry enables precise measurements for sensors, electronics, and biochemical devices. Nonetheless, modeling the electron transfer kinetics of electrochemistry experiments is time-consuming to perform and account for nonideality. We utilize multiple machine learning (ML) approaches to create adaptive and predictive models to rapidly determine the parameters of electron transfer kinetics from heterogeneous experimental square wave voltammograms from surface-bound electrochemistry. Our models include Gaussian process regressions (GPRs), randomized forests of decision trees, and ML ensemble techniques. Whereas conventional approaches with manual background subtraction and analytic modeling took ~10 hours, ML approaches were trained in 0.2-120 minutes and implemented in a fraction of a second. The GPR method ardExponential performed most accurately but exhibited the longest training time. Randomized forests produced respectable estimates with the shortest training times, while ensembles balanced accuracy and time. Adding one to three kinetic parameters improved the training of these ML models with each parameter added. Overall, this shows that ML can efficiently predict kinetic parameters for surface-bound electrochemistry experiments with heterogeneous voltammograms, enabling rapid automated determination of electron transfer kinetics.