Applying Machine Learning to Predict Electron Transfer Kinetics from Voltammetry Experiments.

Area of Science:

• Electrochemistry
• Computational Chemistry
• Materials Science

Background:

• Electrochemical methods are crucial for sensors, electronics, and biochemical devices.
• Modeling electron transfer kinetics in electrochemistry is often time-consuming and complex.
• Surface-bound electrochemistry presents unique challenges for kinetic analysis.

Purpose of the Study:

• To develop rapid and predictive machine learning (ML) models for determining electron transfer kinetics parameters.
• To compare the performance of different ML approaches, including Gaussian Process Regression (GPR), randomized forests, and ensemble techniques.
• To assess the impact of incorporating kinetic parameters on ML model training and predictive accuracy.

Main Methods:

• Utilized heterogeneous experimental square wave voltammograms from surface-bound electrochemistry.
• Developed and trained multiple ML models: Gaussian Process Regressions (GPRs), randomized forests, and ML ensemble techniques.
• Evaluated model performance based on accuracy, training time, and implementation speed compared to conventional methods.

Main Results:

• ML models achieved significantly faster training times (0.2-120 minutes) compared to conventional methods (~10 hours).
• Gaussian Process Regression (GPR) demonstrated the highest accuracy but required the longest training time.
• Randomized forests offered a balance of speed and accuracy, while ensemble methods provided a compromise.
• Incorporating 1-3 kinetic parameters improved ML model training and predictive capabilities.

Conclusions:

• Machine learning provides an efficient and rapid method for predicting kinetic parameters in surface-bound electrochemistry.
• ML enables automated and accelerated determination of electron transfer kinetics from complex electrochemical data.
• The choice of ML model (GPR, random forest, ensemble) can be optimized based on specific accuracy and time constraints.