Estimation of diffusion coefficients from voltammetric signals by support vector and gaussian process regression

Area of Science:

• Electrochemistry
• Computational Chemistry
• Analytical Chemistry

Background:

• Support vector regression (SVR) and Gaussian process regression (GPR) are advanced machine learning techniques.
• These methods were applied to analyze electroanalytical experimental data.
• The primary goal was to estimate diffusion coefficients.

Purpose of the Study:

• To evaluate the accuracy of SVR and GPR in estimating diffusion coefficients from simulated and experimental cyclic voltammograms.
• To compare the performance of SVR and GPR against the conventional Nicholson-Shain equation.
• To investigate the impact of data processing methods on the accuracy of diffusion coefficient estimation.

Main Methods:

• Utilized SVR and GPR with nonlinear kernel/covariance functions for data analysis.
• Simulated cyclic voltammograms based on EC, Eqr, and EqrC mechanisms.
• Compared data reduction techniques: manual peak feature selection, downsampling, and principal component analysis (PCA).
• Applied trained regression algorithms to experimental voltammetric data of organometallic complexes.

Main Results:

• SVR and GPR provided more accurate diffusion coefficients than the Nicholson-Shain equation for simulated voltammograms.
• The accuracy of SVR and GPR was largely independent of reaction rate constants.
• Manual feature selection reduced algorithm performance, while downsampling and PCA were more effective data reduction methods.
• Diffusion coefficients estimated for experimental data closely matched parameter fitting results.

Conclusions:

• Automated voltammogram processing using SVR and GPR offers superior results compared to conventional peak-data analysis.
• These regression algorithms significantly reduce computational time for diffusion coefficient estimation.
• SVR and GPR demonstrate a robust and efficient approach for analyzing electroanalytical data.