Aplikace neuronových sítí a metody ROC v klasifikačních úlohách

The disseratation theses deals with the problem of cost-sensitive binary classification by means of neural networks applied in economical prediction tasks, especially in the field of financial distress prediction. The first part contains the review of existing research in this area and the challenging key points related to cost-sensitive classification are set there. After that, the application of existing Receiver Operating Characteristics (ROC) method, which is able to solve mentioned problems, is discussed and the possibility of its wider use in economical prediction is proposed. The methodology of ROC analysis application is shown in medical and economical experiment of classification with neural networks