科研之友
微信
新浪微博
Facebook
分享链接
摘要
SIMCA (self independent modeling of class analogy) is a classical class modeling method for chemical pattern recognition. Although widely used, SIMCA suffers difficulties in selecting a proper number of principal components and determining the decision region. A new class modeling technique based on partial least squares regression, partial least squares class model (PLSCM) is proposed, where the number of latent variables and decision region can be readily estimated by the routine methods in multivariate calibration, e. g. Monte Carlo cross validation. PLSCM is successfully applied to identify trueborn bezoar samples from artificial and adulterated bezoar samples based on infrared spectra measured in the range of 4000 - 9000 cm(-1). It is demonstrated that PLSCM outperforms SIMCA in terms of both maneuverability and identification accuracy. For the raw spectra, both the training and prediction accuracy of PLSCM are 1.00%. For the standard normal variate-processed data, the training and prediction accuracy of PLSCM is 99% and 100%, respectively.
