Abstract

Support Vector Machine (SVM) is a powerful and flexible learning machine. In recent years combination of SVM and Transductive learning has attracted more and more attention. In many applications such as gene expression, unlabeled data is abundant and available. However labeling data is much more difficult and expensive. Dealing with gene expression datasets, challenges such as curse of dimensionality and insufficient labeled data is inevitable. This paper introduces Iterative Transductive Support Vector Machine (ITSVM). This method which constructs a hyperplane using both training set and working set approximates the optimal solution. Applying proposed algorithm on gene expression datasets show that the proposed method can exploit unlabeled data distribution. In many cases our method improved the accuracy compared to related methods.