Development of a neural network model of credit scoring for physical person

Subject of research: methods and approaches to building neural network models that include processing and analyzing information about borrower's data and are intended for decision-making in the field of lending. Purpose of research: to create a neural network model of a credit scoring system capable of accurately and reliably predicting the creditworthiness of borrowers based on their socio-economic characteristics. Methods and objects of research: the object of research is data on borrowers, including various socio-economic characteristics. The research methods included preliminary data processing, selection of significant features using the SHAP method, and the construction of a fully connected neural network with three hidden layers and a ReLU activation function. The Sigmoid activation function is used for the output layer. The model was optimized using Adam. Main results of research: the developed neural network model was tested on a test sample and demonstrated sufficient accuracy of predictions. The validation showed the minimum value of the loss function at 0.2145 with the optimal number of epochs. The study confirmed the effectiveness of neural network models in credit scoring tasks.