C4.5: programs for machine learning
C4.5: programs for machine learning
Top-Down Induction of Clustering Trees
ICML '98 Proceedings of the Fifteenth International Conference on Machine Learning
Functional bioinformatics for Arabidopsis thaliana
Bioinformatics
Hi-index | 0.00 |
We propose a novel distance based method for phylogenetic tree reconstruction. Our method is based on a conceptual clustering method that extends the well-known decision tree learning approach. It starts from a single cluster and repeatedly splits it into subclusters until all sequences form a different cluster. We assume that a split can be described by referring to particular polymorphic locations, which makes such a divisive method computationally feasible. To define the best split, we use a criterion that is close to Neighbor Joining’s optimization criterion, namely, minimizing total branch length. A thorough experimental evaluation shows that our method yields phylogenetic trees with an accuracy comparable to that of existing methods. Moreover, it has a number of important advantages. First, by listing the polymorphic locations at the internal nodes, it provides an explanation for the resulting tree topology. Second, the top-down tree growing process can be stopped before a complete tree is generated, yielding an efficient gene or protein subfamily identification approach. Third, the resulting trees can be used as classification trees to classify new sequences into subfamilies.