Rediscovering similarity analysis of analytical signal: A not well-known mainstay of chemometrics

Abstract

The search for tools to extract information from analytical signals has led to increasingly complex solutions. Multivariate analysis continues to evolve, exploring new techniques to acquire knowledge. Similarity analysis offers a simple yet powerful tool for comparing analytical signals, aiding in the development of machine learning models for product authentication, quality control, and chemical information mining. When working with 2D signals, like chromatograms or spectra, similarity analysis can be addressed through pairwise or dataset comparisons. Pairwise comparisons use similarity indices, while dataset comparisons typically employ exploratory data analysis. This review highlights the importance of similarity analysis in chemometrics and proposes a classification of different strategies: (i) distance-based approaches, (ii) error-based approaches, (iii) spatial orientation-based approaches, (iv) correlation-based methods, and (v) information entropy-based approaches, aiming to create a consensus and provide analysts with an indispensable tool.