I paradigmi di apprendimento non supervisionato per reti neurali in campo finanziario: progettazione di self-organizing maps per il rintracciamento di anomalie di mercato

The algorithms based on artificial intelligence together with the self-computing numerical methodologies inspired by natural processes are gaining importance in financial issues. The aim of this article is to implement an automatic method for organizing and clustering the information observed on secondary markets, focusing the attention on the recognition of potential anomalies. The methodology, used to highlight potential opportunities of trading, is constituted by a non-supervised neural network, designed to clustering the data, known as Kohonen Network or Self-Organizing Map (SOM). The article is divided into three parts: the first one shows how the methodology works, that is how the structure of the network is defined, how the training procedure is projected and which algorithm must be implemented for a correct mapping of the experimental data. In the second part, we proceed to write and validate the code in a development environment oriented to mathematical modelling; in particular Matlab language has been utilized. In the last part, examples of the applications of SOM in the recognition of anomalies in bond markets are provided.