Classification of Imbalanced Tumor Gene Expression Data Based on Weighted Gaussian Process

Tumor gene expression data is characterized with high dimension, small samples and imbalance. Gaussian Process Classification presents good applicability to high-dimensional and small sample data, but class-imbalanced problem will cause the low classification accuracy on the samples of minority classes. In this paper, a weighted likelihood function is introduced to balance the importance of positive and negative samples during the training of Gaussian Process Classifier, and further extended to multi-class classification tasks. The experimental results on real tumor gene expression data sets show that this method could effectively deal with the class-imbalanced problem caused by One-Against-Rest strategy in multi-class classification, meanwhile, achieve better classification accuracy on both the minority classes and overall data sets.