Multi-class support vector machine based on directed acyclic graph

Support vector machine based on directed acyclic graph (DAG) was proposed for multi-class classification and applied to multi-class fault diagnosis problems. Considering DAG being equivalent to a list operation, and the classification performance depending on the nodes' sequence in the graph, a classification measure based on the distribution of multi-class data was introduced. This method used separability measure between class to estimate distribution character of each class, established the initialization operation list, and organized all sample classes in the list according to certain sequence. The topology structure of DAG based on separability measure was constructed by rearranging the nodes' sequence in the graph. To testify the effectiveness of the proposed method, numerical simulations were conducted on three datasets compared with conventional methods. The results show that, the proposed method has better performance and higher generalization ability.