Journal of Chemometrics

Cover image for Vol. 25 Issue 2

February 2011

Volume 25, Issue 2

Pages 51–99

  1. Research Articles

    1. Top of page
    2. Research Articles
    1. On estimating model complexity and prediction errors in multivariate calibration: generalized resampling by random sample weighting (RSW) (pages 51–58)

      L. Xu, Q.-S. Xu, M. Yang, H.-Z. Zhang, C.-B. Cai, J.-H. Jiang, H.-L. Wu and R.-Q. Yu

      Version of Record online: 18 JUL 2010 | DOI: 10.1002/cem.1323

      A generalized resampling method, random sample weighting (RSW) is proposed to estimate the number of PLS components. Random non-negative weights are assigned to the original training samples and a sample-weighted PLS model is developed without increasing the computational burden much. For prediction, only the training samples with random weights less than a threshold value are selected to ensure that the prediction samples have less influence on training.

    2. Optimisation of epsomite transformation into periclase using experimental design methodology (pages 59–66)

      S. Behij, K. Djebali, H. Hammi, A. H. Hamzaoui and A. M'nif

      Version of Record online: 2 JUL 2010 | DOI: 10.1002/cem.1324

      Experimental design methodology is applied to optimise MgO production. This is a two-step process through an intermediate product (Mg(OH)2). A fractional factorial design and a centred composite one are used to establish experimental conditions for Mg(OH)2 obtention. The favourable conditions for the second step to obtain MgO were determined using a fractional factorial design. The decomposition and crystallisation of Mg(OH)2 was analysed by DTA/TGA and XRD. The morphological properties of the MgO were examined by SEM.

    3. A scalable optimization approach for fitting canonical tensor decompositions (pages 67–86)

      Evrim Acar, Daniel M. Dunlavy and Tamara G. Kolda

      Version of Record online: 27 JAN 2011 | DOI: 10.1002/cem.1335

      Tensor decompositions are higher-order analogues of matrix decompositions and have proven to be powerful tools for data analysis. For fitting the CANDECOMP/PARAFAC tensor decomposition, we propose the use of gradient-based optimization methods such as nonlinear conjugate gradients. Computational experiments demonstrate that the gradient-based optimization methods are more accurate than the standard alternating least-squares (ALS) and faster than second-order optimizaion in terms of total computation time.

    4. Relation between second and third geometric–arithmetic indices of trees (pages 87–91)

      Boris Furtula and Ivan Gutman

      Version of Record online: 8 OCT 2010 | DOI: 10.1002/cem.1342

      In this paper, the peculiar relation between second and third geometric–arithmetic indices of trees was investigated. It was found that it depends on the number of pendent vertices. In addition, the lower and upper bounds of GA3 index of trees in terms of GA2 index and number of pendent vertices, ν, was derived.

    5. Combination of kernel PCA and linear support vector machine for modeling a nonlinear relationship between bioactivity and molecular descriptors (pages 92–99)

      Guang-Hui Fu, Dong-Sheng Cao, Qing-Song Xu, Hong-Dong Li and Yi-Zeng Liang

      Version of Record online: 1 FEB 2011 | DOI: 10.1002/cem.1364

      In this paper, a two-step nonlinear classification algorithm is proposed to model the structure-activity relationship (SAR) between bioactivities and molecular descriptors of compounds, which consists of kernel principal component analysis (KPCA) and linear support vector machines (KPCA+ LSVM). The combination of KPCA and LSVM can effectively improve the prediction performance compared with the linear SVM as well as two nonlinear methods. Three datasets related to different categorical bioactivities of compounds are used to evaluate the performance of KPCA+LSVM. The results show that our algorithm is competitive.