Management of Semiconductor Manufacture--A Discussion on Multi-class Classification of Imbalanced Structure of IC Package Database

In the past, for the imbalance class distribution, in most cases the representative class data were chosen by sampling, in order to improve the efficacy of the class distribution model in predicting the minority of classes in the imbalanced data set. The research attempts to present a new pre-processing method of data - the Orthogonal Transformation Method (OTM), which, by integrating the conceptions of Taguchi Orthogonal Arrays, without changing the original data structure, improves the Orthogonality of the data structure by adding variables so that the accuracy of the automatic class distribution database of IC products of imbalanced data set is improved, the range of information retrieval is accurately narrowed, the efficiency and the quality of retrieval can be exalted to a great extent and thus the performance of IC design is upgraded. For the first year, the programs to be implemented and expected results are: Orthogonal Transformation Method, programming and performance evaluation.