Advanced Methods for the Analysis of Semiconductor Manufacturing Process Data - Advanced Techniques in Knowledge Discovery and Data Mining - page 66

Fig. 2.23. Visualization of selected SPLIT3 according to compactness q ci .
After the regarded steps of interactive visualization, analysis, and au-
tomatic determination of relevant variables, the monitoring of the process
state by classification methods is investigated. According to the underlying
lots, SPLIT6 was separated after feature selection (SBS, q si , 9 features) into
a training set, SPLITTrain3, and a test set, SPLITTest3. The six different
classes in SPLIT6 were due to the distinguishing of the lots. Splitting SPLIT6
into a training and a test set reduces the classification task to an L =3class
problem. In the first step of this part of the work, the training set was used
to train a reduced nearest neighbor classifier (RNN) [2.13]. As can be seen
from Fig. 2.24, generalization was perfect and data from the second lot can
perfectly be classified according to the three split classes and the features
chosen for optimum separability. However, in this approach numerous sam-
ples of the novel or abnormal cases were available. In the second step of this
part of the work, OCC was applied to the same data. It must be kept in mind