A proposed method of local feature-weighting to improve predictions of basic nearest neighbor rule
ASC '07 Proceedings of The Eleventh IASTED International Conference on Artificial Intelligence and Soft Computing
Hi-index | 0.00 |
High dimensionality is one the major problem in the classification of microarray gene expression data. Most of the classifiers performed well for the data having same number of features as the number of samples. But gene expression data have very few samples as compare to the number of genes or features. We use Class Prediction (CP) with Compound Covariant Predictor (CCP), Diagonal Linear Discriminant Analysis (DLDA), k-Nearest Neighbor (NN), Nearest Centroid (NC) and Support Vector Machine (SVM) to create multivariate predictor to determine the class of a given data sample. In this paper, CP has been used to classify the tumor groups from the microarrays dataset taken from breast cancer patients. The paper presents comparative results to determine the accuracy of a cancer gene classification based on six multivariate classifiers. Our results have shown that CCP has performed best with an accuracy of 100%, 85% and 86% among three tumor groups. Accurate analysis and classification of gene expression profiles could lead to more reliable tumor classification, better prognostic prediction and selection of more appropriate treatments.