Application of Bayesian Networks in Knowledge Mining

The thesis focuses on the application of Bayesian networks in credit risk modeling within the banking sector, where both predictive accuracy and model interpretability play a crucial role. Bayesian networks, based on Bayes' theorem and conditional probabilities, enable efficient modeling of relationships between variables and provide well-interpretable outputs, which is essential in a regulated environment. The study compares the predictive capabilities of Bayesian networks with logistic regression and the Naive Bayes classifier, using evaluation metrics such as accuracy and AUC. Emphasis is placed on comparing Bayesian networks with traditional methods in assessing the probability of default. The practical part of the thesis includes the design and implementation of a model that assists managers in identifying high-risk clients and compares its outcomes with those of standard models. The results demonstrate that Bayesian networks represent an attractive alternative, offering a combination of predictive accuracy and interpretability. This thesis also includes a decision support system for assessing clients' credit risk.