Granular SVM with Repetitive Undersampling for Highly Imbalanced Protein Homology Prediction
Yuchun Tang, Yanqing Zhang · 2006
Highly imbalanced classification is important and increasingly common with emergence of new machine learning application domains including biomedical informatics. In order to solve this challenging class imbalance problem, a novel Granular Support Vector Machines - Repetitive Undersampling algorithm (GSVM-RU) is designed in this work. GSVM-RU creatively utilizes Support Vector Machines (SVM) themselves for undersampling to minimize the negative effect of information loss while maximizing the positive effect of data cleaning in the undersampling process. Consequently, an accurate and fast classifier can be modeled. GSVM-RU ranks as one of the best solutions in ACM KDDCUP 2004 competition for the extremely imbalanced protein homology prediction.