Unsupervised Analysis

Unsupervised analysis is performed when there are no other samples, information or data than the sample itself and the Raman map to be analyzed. When the sample is truly unknown, unsupervised analysis is the first (sometimes only) step in characterizing the sample. The objective of the analysis would be to determine the chemical and spahal compositions of the unknown sample. 

Raman spectroscopy, compared to IR spectroscopy, is considered to be easy spectroscopy. Often, it is possible to identify spectral features that are unique to each chemical component in the sample, and to use them for univariate analysis. When the number of spectra in a Raman map is small and the chemical composition of the sample is simple, then a univariate analysis is usually sufficient. The typical analysis strategy would be to pretreat the spectra (baseline correction, nor-malizahon, etc.), explore to identify any unique spectral features, and to create intensity maps of those features as Raman images. When differences between spectral species are reflected in peak shifts or band broadenings, the peak positions or bandwidths can also be mapped to create Raman images. 

When performing an unsupervised multivariate analysis it is important to remember that it is a numerical analysis based on the variance of the data set Whenever the data set is manipulated, the results may change. Hence, it is imperative to verify the results with the original spectra to confirm the chemistry. Repetition is also a form of variance when there is more repetition of a spectral species, the loading will seem smoother. Conversely, even if the SNRs of the original spectra are similar the loading of a spectral species of less repehtion will seem noisier. 

Figu re 11.10 Histograms of (a) caffeine and (b) cellulose score images. The x-axis shows the score the y-axis the number of spectra. 

This is one of the best examples where all criteria for a good approximahon (good spahal and spectral segregation between components, similar density of all components, etc.) are met, and demonstrates the potential usefulness of the approach. 

---

Verification is confirmation by examination and provision of objective evidence that specified requirements have been fulfilled. ISO 9000 2005 Proficiency testing is a periodic assessment of the performance of individual laboratories and groups of laboratories that is achieved by the distribution by an independent testing body of typical materials for unsupervised analysis by the participants. [lUPAC Orange Book]... 

Pattern Recognition I Unsupervised Analysis For the test data,... 

Figure 3.41 shows the result of imaging mass spectrometry-principal component analysis (IMS-PCA) for the colon cancer tissue. In this case, this unsupervised analysis revealed that the largest spectral difference (i.e., the largest difference in metabolite composition) was observed between the normal and the other tissue areas (i.e., normal vs. stroma/cancer area), and the second largest difference was observed between the stroma and normal/cancer area. The overall interpretation of PCA was shown in Table 3.6. 

The main purposes of multivariate analysis are data reduction (unsupervised analysis) and data modeling like regression and/or classification models (supervised analysis). 

---