Improvement on enhanced Monte-Carlo outlier detection method

Highly predictive multivariate calibration model depends on samples in training set. In this study, we introduced an outlier detection method and developed its improvement for shorter run time. Improved Monte-Carlo outlier detection (IMCOD) was proposed to establish cross-prediction models for deter...

Full description

Saved in:
Bibliographic Details
Published in:Chemometrics and intelligent laboratory systems 2016-02, Vol.151, p.89-94
Main Authors: Zhang, Liangxiao, Wang, Du, Gao, Rongrong, Li, Peiwu, Zhang, Wen, Mao, Jin, Yu, Li, Ding, Xiaoxia, Zhang, Qi
Format: Article
Language:English
Subjects:
Monte-Carlo sampling
Multivariate calibration
Outlier detection
Partial least squares regression
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Highly predictive multivariate calibration model depends on samples in training set. In this study, we introduced an outlier detection method and developed its improvement for shorter run time. Improved Monte-Carlo outlier detection (IMCOD) was proposed to establish cross-prediction models for determining normal samples, which were subsequently used to analyze the distribution of prediction errors for all of dubious samples together. Four real datasets were employed to illustrate and validate the performance of IMCOD. After sample selection for training set of NIR of soy flour samples, the Root Mean Square Error of Prediction (RMSEP) of PLS model decreased from 1.4811 to 0.7650. This method benefits the establishment of a good model for QSAR and NIR datasets. •IMCOD was proposed to detect outliers based on Monte Carlo sampling.•IMCOD could overcome masking effect by taking dubious samples as test set.•The performance of IMCOD outperforms MCOD and EMOCD.
ISSN:0169-7439
1873-3239
DOI:10.1016/j.chemolab.2015.12.006