Pattern recognition system based on empirical knowledge: Sandalwood and camphoraceous odors application
Résumé
In the first part of this paper, structure-odour relationships for sandalwood odourants were studied for a set of 158 compounds (75 inactive and 83 active). The first step in the study consists in an evaluation of empirical rules concerning structure-sandalwood odour relationships, by means of decision trees. The main structural elements responsible for other than sandalwood odours are defined from the empirical rules. A search for new variables was carried out to define the essential elements for the sandalwood odour. The rules used to discriminate between sandalwood and non-sandalwood molecules lead to 80% correct discrimination. In the second part, fuzzy logic has been used as a tool in structure-camphoraceous odour relationships. The database studied included 99 molecules. The rules used to discriminate between camphor and non-camphor are given by an expert. Such rules account for the shape and the size of the molecules. Their adjustment by means of genetic algorithms (GAs) lead to 84% correct discrimination between camphor and non-camphor molecules.