Approaches to risk analysis in the financial sector based on machine learning and artificial intelligence methods

The article studies approaches to improving the forecasting quality of machine learning models in finance. An overview of studies devoted to the application of machine learning models and artificial intelligence in the banking sector is given, both from the point of view of risk management and considering in more detail the applied methods of credit scoring and fraud detection. Aspects of applying explainable artificial intelligence (XAI) methods in financial organizations are considered. To identify the most effective machine learning models, the authors conducted experiments to compare 8 classification models used in the financial sector. The gradient boosting model CatboostClassifier was chosen as the base model. A comparison was carried out for the results obtained on the CatboostClassifier model with the characteristics of the other models: IsolationForest, feature ranking model using Recursive Feature Elimination (RFE), XAI Shapley values method, positive class weight increase models wrapper model. All models were applied to 5 open financial data sets. 1 dataset contains transaction data of credit card transactions, 3 datasets contain data on retail lending, and 1 dataset contains data on consumer lending. Our calculations revealed slight improvement for the models IsolationForest and wrapper model in comparison with the base CatboostClassifier model in terms of ROC_AUC for loan defaults data.