An outline of proposed research directions and the desirable network configurations, infrastructure and facilities to support research in chemometrics

Exploratory Data Analysis: Patterns may exist in many data sets, but can be difficult to discover in a large, multivariate, complex data. Exploratory data analysis can reveal hidden patterns by reducing the information to a more comprehensible form. A chemometric analysis can indicate patterns and trends in the data. Usual techniques employed include principal component analysis and hierarchical cluster analysis that reduce the dimensionality of the data. These techniques can emphasize the natural groupings in the data and indicate which variables most strongly define the patterns. As an integral part of the pattern recognition process, exploratory data analysis is often followed by multivariate classification.

Classification Modeling: Many analytical applications require that samples be assigned to predefined categories, or "classes", i.e. perform a diagnosis. This may involve determining whether a sample is good or bad, or predicting an unknown sample as belonging to one of several distinct groups. A classification model is used to predict a sample's class by comparing the sample to a previously analyzed set, in which categories are already known. There is considerable interest worldwide in developing and applying pattern recognition tools to biospectroscopic analysis. Interrogation of infrared, and more recently proton NMR, databases is providing a rich field of new applications for these techniques.

Regression Analysis and Quantification: In many applications, it is expensive, time consuming or difficult to measure directly a property of interest. The analyst is required undertake the prediction based on related properties that are easier to measure. The goal of chemometric regression analysis is to develop a calibration model which correlates the information in the set of known measurements to the desired property. Chemometric algorithms for performing regression include partial least squares and principal component regression and are designed to avoid problems associated with noise and correlations in the data. These methods are often suitable for on-line monitoring and process control, where fast and inexpensive systems are needed to test, predict and make decisions about product quality. New approaches to allow continuous update of multivariate calibrations in process analysis continue to be researched, along with the search for methods to allow the transfer of calibration models between instruments. The development of calibration models siuitable for smart sensors will become more important.

ARNAS would provide the opportunity for scientists in Australia to contribute to, and benefit from, the interdisciplinary teams in around the world.

[ Back ] [ Up ] [ Next ]