Application of Global Model for Classification Across Different Experiment Studies


A new modelling approach for classification across different experimental studies is described in this paper. It involves extracting common components present in different experiments to form a global model for cross experimental study and a local model being constructed for comparison. In order to reduce the dimensionality of models, Principal Component Analysis (PCA) is applied to construct Principal Component (PC) models, the number of PCs being optimised by the ratio of the Predicted Residual Error Sum of Squares (PRESS) to the Residual Sum of Squares (RSS) using 200 bootstrap repetitions. The multi-class Quadratic Discriminant Analysis (QDA) classifier is applied for different two-class classifications, the assignment of samples depending upon the smallest distance of the samples to the centroid of the classes. Model performance is assessed through 100 test sets using different iterative splits and two performance indices. Compared to the local model consisting of the same samples, the global model displays better performance and most stress samples can be correctly assigned to their own class.


全文下載 Acrobat PDF

Joomla Templates by