KÖKSAL G. (Executive)
TUBITAK Project, 2006 - 2009
The objective of
this project is to identify the data mining (DM) approaches that can
effectively improve product and process quality in industrial organizations,
and to develop more effective approaches. In the project, quality definition,
prediction, classification and parameter optimization problems associated with
product and process quality improvement in manufacturing industries are
considered. For the solution of these problems, clustering, prediction,
classification, association and optimization functions of DM as well as data
preparation and preprocessing are determined as relevant. A comprehensive
literature survey has been performed and six manufacturing companies operating
in different sectors have been visited, within this context. Appropriate DM
methods are applied on data sets obtained from three of these companies, and the
results are compared. As a result, the most appropriate DM methods are
suggested for specific DM functions and quality improvement purposes. In the
method development part of the project, studies are performed to overcome some
problems encountered during the applications, and to increase ease of use and
effectiveness of the VM methods. As a result, a resampling method for quality
data; an alternative nonparametric approach (CMARS) for regression; adaptations
of an easy to use binary classification method, Mahalanobis Taguchi system, to multiple
classes and also to parameter optimization; alternative approaches for fuzzy
classification of quality data (models based on fuzzy regression) and
nonparametric fuzzy functions; alternative approaches for optimization of
desirability functions in parameter optimization; and a method for reduction of
association rules are developed. It is expected that these results and
approaches guide practitioners in quality improvement area, and increase the
ease of use and effectiveness of them.