Greedy hierarchical binary classifiers for multi-class classification of biological data

Abstract

Multi-class classification is an important and challenging problem for biological data classification. Typical methods for dealing with multi-class classification use a powerful single classifier such as neural networks to classify the data into one of many classes. Alternatively, the binary classifiers are used in one-versus-one (OVO) and one-versus-all (OVA) classifier schemes for multi-class classification. However, it is not clear whether OVO or OVA yields good performance results. In this paper, we propose a greedy method for developing a hierarchical classifier where each node corresponds to a binary classifier. The advantage of our greedy hierarchical classifier is that at the nodes any type of classifier can be used. In this paper, we analyze the performance of the proposed technique using neural networks and naive Bayesian classifiers and compare our results with OVO, OVA, and exhaustive methods. Our greedy technique provided better and more robust accuracy than others in general for biological data sets including 3- to 8-classes.