Java Reference
In-Depth Information
particular supplier's parts or even the repair process itself. Using
attribute importance, we can identify the factors that best determine
valuable and problem parties. Using association, it is possible to
determine which types of problems typically occur together and
from which parties. By identifying the causes of failures faster,
overall costs can be reduced by taking corrective action sooner. In
addition, data mining can be used to forecast warranty reserves.
Warranty claims processing is often labor intensive, especially
where invalid or fraudulent claims must be analyzed manually. By
quickly classifying claims as invalid or potentially fraudulent, using
techniques discussed in Section 2.1.4, both time and claims paid can
be reduced.
2.1.10
Defect Analysis
Any company that manufactures products is interested in under-
standing the root causes of product defects. Data mining provides
several techniques for uncovering root causes, including association,
clustering, and predictive modeling. In association, the resulting
rules identify co-occurring items, features, or events that typically
accompany defects. Whereas the number of factors associated with
each production run may be large, on the order of hundreds or
thousands of factors, such that examining these manually is intracta-
ble, the association data mining technique extracts rules highlighting
the most frequent factors that, in this case, result in product failure.
In clustering, defective parts or production runs may share
common properties. Clustering analysis produces groupings of items
according to common attribute values. Reviewing the distribution of
data values for each cluster and the rules that define each cluster may
give clues as to what each of these defects has in common. In predic-
tive modeling, one can build a classifier to predict which items are
likely to be defective. If the classification model provides transparency,
such as the rules, one can identify what determines a failure. More-
over, for any given production run, users could predict if a defect is
likely given certain conditions such as temperature or humidity. Other
benefits of data mining for defect analysis are well-stated in [Wu 2002]:
Through the use of data mining techniques, manufacturers are able to identify
the characteristics surrounding defective products, such as day of week and time of
the manufacturing run, components being used and individuals working on the
assembling line. By understanding these characteristics, changes can be made to
the manufacturing process to improve the quality of the products being produced.
Search WWH ::




Custom Search