Study on the Outliers Recognition Method for the Chemistry and Chemical Data
Abstract: In the field of the chemistry and chemical engineering, due to the incomplete of data collection equipment, data input error, error transfers, confusion measure unit, inadequate delicacy, handcraft import course lose data and so on, outliers often be brought. The outliers often bring out the whole information application kickback. So in the article, the partial least squares (PLS) method based on the distance was selected to study and recognize the outliers from the whole sample data. The PLS method derived several components from the independent variables and thus eliminated spectrum data the severity collinearity. Thereby the PLS method brought the low dimension model and the model became concise. Finally, the proposed PLS method application to the analysis of fish data near infrared spectroscopy which dealt with spectrum data to get along with ration forecast model of fatness content . The result of the PLS method was presented with comparison to the MLR. The PLS method not only gets the high right ratio of the outlier, but also holds lower ratio of the right samples.
The paper has also analyzed the research work insufficiency, and has prospected the next development of the outliers recognition method
Keywords：Outliers; Partial least squares; Multiple Linear Regression; NIR
