Classification Of Ukrainian Banks Using Machine Learning Methods In Martial Arts Conditions

Under wartime conditions, Ukraine's banking system underwent significant structural transformation: bank differentiation by profitability, capitalisation and operational efficiency intensified, while asset structures shifted toward risk-free government securities. Traditional cluster analysis identifies typological groups but lacks a mechanism for validating results on new observations, necessitating its integration with machine learning methods.The purpose of the study is to develop and validate a two-stage methodological system for classifying Ukrainian banks by combining cluster analysis and machine learning to identify stable typological groups and assess classification accuracy under martial law.The study draws on NBU supervisory data for 60 banks. Five indicators are used: return on assets, capital adequacy ratio, cost-to-income ratio, reserves-to-loans ratio and the natural logarithm of total assets. Stage one applies k-means clustering; stage two employs PLS discriminant analysis with ten-fold cross-validation and an 80/20 train–test split.A two-stage classification system combining k-means and PLS discriminant analysis was developed and validated, yielding a machine learning model with 87.5% accuracy suitable for regulatory monitoring.The PLS model enables the NBU to detect banks migrating toward the inefficient cluster for pre-emptive supervisory action. Future research directions include ensemble machine learning methods and panel analysis covering 2022–2025.