Chem om etric Analysis of D iffuse Re ̄ ectance - Absorbance Fourier Transform Infrared Spectra Using Rule Induction M ethods : Application to the C lassi ® cation of Eubacterium Species

Strains represen tative o f fou r Eubacterium species were analyzed by u sin g d iffuse re ̄ ectan ce-absorbance Fourier transform infrared spectroscopy. To iden tify im portan t wavenumber reg ions for the classi ® cation of th ese 22 bacterial isolates, we invest igated three ru le induction m ethod s and variou s spectral preprocessin g regim es. In th is stu dy both univariate and m ultivariate classi ® cation and regression trees (CART) m ethod s and the fuzzy m ultivariate rulebu ilding expert system (FuRES ) m ethod were exploited . It was found that the FuRES method was superior in term s of p red iction , whereas the rules proposed by the univariate CART method were easier to interpret in term s of which wavenumbers in th e IR spectra were importan t for bacterial class separation. S caled and detrended FT-IR spectra and ® rst-o rder numerical d ifferen tia tion preprocessing steps were necessary to obtain optim al classi ® cation models. F inally, a reduction in th e classi ® cation error for the CART-based m ethod s was observed by analyzing the com pressed B-sp line coef® c ien ts rather than the original spectra representation . The spectral in terpreta tion of these rules is in agreem ent w ith ana lyses using th e uncompressed representation .