| چکیده انگلیسی مقاله |
Recognizing outlier data and discarding them is one of the steps in data processing that has been considered in various sciences. The existence of outliers in chemistry data is not unexpected [1]. Regression for calibration and prediction of analyte concentration can be affected by outlier data. In performing the calibration, if the outliers are not identified and discarded, the constructed model will not have much validity and we will also make mistakes in estimating the concentration of unknown samples. [2]. In this work, the data will be divided into two categories, univariate and multivariate, and in the case of univariate data, which we will encounter in most analytical techniques, we will extract a map that recognizes the following: a) large error in the experimental process: the recorded signal does not match the desired concentration. B) The existence of data with high regression coverage that is far from the rest and will strongly affect the accuracy of the calibration equation, high leverage data. C) Detection of data with a signal less than the standard deviation of regression, the concentration of which cannot be predicted by the extracted equation. The proposed roadmap on various analytical techniques will be reviewed and its efficiency studied, and this will give a new look at the calibration equation, how to distinguish the upper and lower limits of the equation, and a solution to avoid common mistakes in chemistry. In the case of multivariate data [3], we will extract an outlier map similar to the univariate data outliers map, also has another special ability; detection of the presence of a non-calibrated interference in unknown samples, first-order data advantage. These methods will be implemented for different types of multivariate calibration data and models. |
