| Author | Message | ||
     Julio Trevisan (lascanter2010) Member Username: lascanter2010 Post Number: 13 Registered: 8-2010 |
Hi, I am working on a project where I am comparing different data handling methodologies following acquisition of spectra using ATR-FTIR spectroscopy. At this point of the project, I have addressed many things. However, I really feel that I need the consulting advice a spectroscopy specialist. Specifically, I want to look at the influence of different types of pre-processing in both the performance of a classification system and in the biomarkers (important wavenumbers) that are read out of this system. From a pure machine learning point of view, I would choose the pre-processing method for which my classifier gives best performance. However, I read in the literature that the choice pre-processing methods is usually justified to the type of spectroscopy and instrument. Going further, I noticed that different pre-processing methods give me different biomarkers (wavenumbers selected in the feature selection stage). In my project, gaining information about the biological problem is important. However, it seems that pre-processing the data differently may give rise to a different biological interpretation, because of different wavenumbers picked. I don't know if this discussion is interesting, or maybe it has been addressed before?, or there is an easy way out that would make it pointless? If anyone has any experience in addressing these questions, any advice would be greatly appreciated. Julio Trevisan PhD Student Lancaster University, UK |