This thesis deals with a decision tree approach for analyzing the process monitoring data. Decision tree is a data mining technique and has been widely used for the purpose of classification. Decision tree is also useful in analyzing the training data which may contain disturbances.
The existing data mining techniques conduct a binary split of a continuous-valued attribute for discretization. This form of classification is very fast and easy to interpret. However, the binary split has a problem in analyzing process monitoring data in that it is difficult to determine process parameter`s upper and lower specification limit at the same time.
In addition, usual decision tree analysis techniques have focused on a single performance characteristic. However, multiple performance characteristics appear more commonly in the process monitoring data. Therefore, a new approach is necessary to analyze multiple performance characteristics simultaneously.
In this thesis, a decision tree technique with ternary split at each node is developed for finding principal process parameters which influence the performance characteristic. The developed method can be used to determine the specification limits of the process parameters in both cases of single and multiple performance characteristics.
