A method for discovering data patterns through constructing feature networks

With the arrival of the big data era, data is playing a prominent role increasingly. At present, data is stored in a variety of ways, which relational data is one of the most important. This paper aims to combine the method of data mining and complex network to analyze and utilize relational data, and a method of constructing feature network is proposed to discover some interesting patterns hidden in the massive relational data. First, a method is introduced to transform relational data to complex networks, in which the features of data is defined as nodes of networks and the correlation of two features is taken as edge weight of networks. Second, some measures of feature networks is calculated to find some data patterns. Finally, the above method is applied in two medical data sets, and the analyzed result shows that different kind of data (healthy people and patients) is significantly different in their topology of feature networks. The method is expected to provide an efficient way to discover data patterns and classifying data.