Self-learning scoring models – introduction of an on-line approach to risk assesment

The problem considered in this article involves the construction of evaluation model, which could subsequently be used in the field of modeling and risk management. The research work is finalized by a construction of a new model on the basis of observations of the models used for risk management and knowledge of information theory, machine learning and artificial neural networks. The developed tools are trained online, using their ability for automatic deduction rules based on data, during model application for evaluation tasks. The model, consequently changes the data analysis stage, limits the scope of the necessary expertise in the area, where the assessment model can be used and, to some extent, the shape of the model becomes independent from the current range of available data. These features increase its ability to generalize and to cope with the data of previously undefined classes, as well as improve its resistance to gaps occurring in the data. Performance of the model presented in this paper is tested and verified on the basis of real-life data, which would resemble a potentially real practical application. Preliminary tests performed within the scope of this work indicate that the developed model can form a starting point for further research as some of the used mechanisms have a fairly high efficiency and flexibility.