Weightless neural networks on ranking of financial client solvency

Boolean neural networks are known for their low cost and ease of hardware implementation. Networks of goal seeking neurons (GSN) encompasses an interesting class and were designed to solve problems from networks of probabilistic logic neurons (PLN), successors of the very first Boolean model. The use of neural networks in financial applications is being consolidated through the last decade. This paper proposes a GSN based system to cope with ranking problems. We are concerned with the ranking of clients with respect to financial solvency. On the contrary of more popular models, GSN networks have a one-shot learning algorithm (where each example is shown only once) combined with the nondestruction of acquired knowledge. These features lead to inability to learn nonrepresentative patterns and show it is possible to define the similarity degree of a client with respect to the whole class. The proposed system aims, therefore, to define how solvent and insolvent a client is in relation to the universe of classified clients presented to the network. The proposed system also implements earlier research on the treatment of false examples by GSN control of presentation order