A New Data Mining Method of Iterative Dimensionality Reduction Derived from Partial Least-Squares Regression

The main information retrieval and information noise elimination were the essential technology for data mining. The multiple correlation among multi-index was one of the main reasons for difficult to determine independent variables set in a data mining regression. The paper introduced a new iterative dimensionality reduction method based on partial least-squares regression. Most of the independent variables set should be contained in the original mathematical model in order to avoid missing necessary important information. Some irrelevant or less relevant variables were excluded through successive iterations and the conditions ensuring model accuracy and minimizing the loss of information must be matched at the same time. Ultimately the regression model including important variables set was highly refined. The truly physical non-linear model reflected the relationship among magnetic field strength, strain and magnetic frequency in giant magnetostrictive material (GMM) was deduced by using the iterative method.