Improving Regression Models by Dissimilarity Representation of Bio-chemical Data
The determination of characteristics by regression models using bio-chemical data from analytical techniques such as Near Infrared Spectrometry and Nuclear Magnetic Resonance is a common activity within the recognition of substances and their chemical-physical properties. The data obtained from the mentioned techniques are commonly represented as vectors, which ignore the continuous nature of data and the correlation between variables. This fact affects the regression modeling and calibration processes. For solving these problems, alternative representations of data have been previously used with good results, such as those ones based on functions and the others based on dissimilarity representation. By using the alternative based on dissimilarities, the obtained results improve the efficiency of the classification processes, but the experience in regression with this representation is scarce. For this reason, in this paper, in order to improve the quality of the regression models, we combine the dissimilarity representation with some adequate data pre-processing, in our case, we use the classical Partial Least Square regression as the modeling method. The evaluation of the results was carried out by using the coefficient of determination \(R^2\) for each case and a statistical analysis of them is performed.
KeywordsDissimilarity representation Regression Bio-chemical data
- 3.Esbensen, K.H., Guyot, D., Westad, F., Houmoller, L.P.: Multivariate data analysis: in practice: an introduction to multivariate data analysis and experimental design. In: Multivariate Data Analysis (2002)Google Scholar
- 6.Massart, D.L.: Handbook of Chemometrics and Qualimetrics. Elsevier Science, Amsterdam (1997)Google Scholar
- 8.Porro Munoz, D.: Classification of continuous multi-way data via dissimilarity representation (2013)Google Scholar
- 11.Thodberg, H.: Statlib-datasets archive website (2018). http://lib.stat.cmu.edu/datasets/tecator