Decision Network: a New Network-Based Classifier

Abstract

In recent year, the combination of machine learning and complex networks is gaining more and more attention. Some network-based machine learning methods which transform the vector-based instances into a network has shown a lot of potential. Some researchers believe that the network can show more information than vector-based datasets. In this paper, we proposed a network-based classifier named decision network(DN). DN abstracts the corresponding relationships between attribute values and class labels into a weighted bipartite network. The weight of the edge between an attribute value node and a label node represents the tendency to assign the instance with this attribute value to the corresponding class. Compared with the existing classifier, DN is more comprehensible and easier to implement. We evaluated the performance of DN on 7 real-world datasets by using 10-fold cross validation. It performs better than 9 other methods.