Citation:
Proceedings of the 14th International Symposium on Olfaction and Electronic Nose, New York City, NY, USA, 2-5 May 2011, Pages 9-10. Editor: Perena Gouma.
Abstract:
Pattern recognition is essential for translating complex olfactory sensor responses into simple outputs that are relevant to users. Many approaches to pattern recognition have been applied in this field, including multivariate statistics (e.g. discriminant analysis), artificial neural networks (ANNs) and support vector machines (SVMs). Reviewing our experience of using these techniques with many different sensor systems reveals some useful insights. Most importantly, it is clear beyond any doubt that the quantity and selection of samples used to train and test a pattern recognition system are by far the most important factors in ensuring it performs as accurately and reliably as possible. Here we present evidence for this assertion and make suggestions for best practice based on these findings.
