Unsupervised classification of voltammetric data beyond principal component analysis.

Area of Science:

• Electrochemistry
• Machine Learning
• Data Science

Background:

• Cyclic voltammetry (CV) is a crucial electrochemical technique.
• Unsupervised classification of CV data is challenging due to its complexity.
• Developing robust classification methods is essential for electrochemical analysis.

Purpose of the Study:

• To evaluate various unsupervised classification approaches for cyclic voltammetric data.
• To compare the efficacy of Principal Component Analysis (PCA), t-distributed Stochastic Neighbour Embedding (t-SNE), and Uniform Manifold Approximation and Projection (UMAP).
• To investigate the use of transfer learning with VGG-16 for feature extraction in classification.

Main Methods:

• Applied unsupervised learning algorithms: PCA, t-SNE, and UMAP.
• Utilized transfer learning with the VGG-16 image recognition network for feature extraction.
• Tested methods on both noise-free and noisy cyclic voltammetric datasets.

Main Results:

• t-distributed Stochastic Neighbour Embedding (t-SNE) exhibited the best performance in classifying CV data.
• UMAP and PCA showed comparatively lower performance.
• Transfer learning with VGG-16 features improved classification accuracy, especially in the presence of noise.

Conclusions:

• t-SNE is the most effective unsupervised method for cyclic voltammetric data classification.
• Feature extraction via VGG-16 combined with dimensionality reduction techniques enhances classification robustness.
• The findings provide valuable insights for electrochemical data analysis and interpretation.