Mining complex networks: A new challenge for supporting diagnostic decisions

New technologies are multiplying at an enormous speed and the produced data is not only massive but also complex. In fact, despite the abundance of tools to capture, process and share information (e.g. data) one cannot broadly assume the standard hypothesis that such data are identically and independently distributed (i.i.d.). As a result, proper handling of data is fundamental in order to convert the available observation in to useful information that leads to knowledge and suitable decision making. In this paper, we focus on network data. That is, we introduce the reader to a theoretical perspective concerning the knowledge mining of huge amount of relational information collected in all the network systems which are ubiquitous in our life. In this context, following a numerical evaluation we show the reader how different kind of information can provide a benefit for a typical machine learning problem i.e. classification. The main issue of our investigation is to provide a case where the accuracy of a classification model benefits when considering the additional information given by both network and dissimilarity features. Moreover, we treat a clinical example that will serve as running case for our analysis.