Pattern recognition applied to vapour-phase infrared spectra

✍ Scribed by L. Domokos; I. Frank; G. Matolcsy; G. Jalsovszky

• Publisher: Elsevier Science
• Year: 1983
• Tongue: English
• Weight: 573 KB
• Volume: 154
• Category: Article
• ISSN: 0003-2670
• DOI: 10.1016/0003-2670(83)80018-x

✦ Synopsis

Various supervised and unsupervised pattern recognition methods are compared, to study the effect of data reduction and preprocessing scale transformations on recognition ability. Data were taken from a collection of 2300 vapour-phase infrared spectra. The basis of chemical classification was the presence or absence of carboxy, amino, alcoholic hydroxy, nitro, ester and phenyl groups, respectively.

Pattern recognition methods, such as binary linear separation or learning machine [l-7], KNN [6, hierarchal clustering [ll, 121 and maximum likelihood [6], have been applied to the interpretation of infrared spectra, to separate chemical classes of organic compounds. The aim of the present study was to compare the performance of various supervised and unsupervised pattern recognition methods and to examine the effect of data reduction and preprocessing scale transformations on recognition ability. Because vapour-phase infrared data were used the expectation was that, in the absence of intermolecular interactions, the separation of categories would be better than in previous investigations.