Biased locality-sensitive support vector machine based on density for positive and unlabeled examples learning

Learning from positive and unlabeled examples (PU learning) has been a hot topic for classification in machine learning. The key feature of this problem is that there is no labeled negative training data, which makes the traditional classification techniques inapplicable. According to this feature, we propose an algorithm called biased locality-sensitive support vector machine based on density (BLSBD-SVM) for PU learning which takes unlabeled examples as negative examples with noise. Our approach as the variant of Locality-Sensitive support vector machine (LSSVM) not only has a lot of advantages of local learning, but also makes good use of the prior information of training examples by adding the relative density degrees of training points. The experimental results on bioinformatics data show the effectiveness of our algorithm.