Abstract
This study aims to enhance the prediction of bank financial unsoundness by develop-ing an integrated model that combines multiple discriminant analysis (MDA), logit, and probit approaches for Kazakhstan’s banks. Using data from 12 banks between 2008 and 2012, the models were tested on 2013-2014 data, incorporating financial ra-tios that reflect capital adequacy, asset quality, management, earnings, and liquidity. The MDA, logit, and probit models achieved predictive accuracies of 83.3%, 87.5%, and 83.3%, respectively, with Type I errors at 8.3% and Type II errors between 16.7%-25%. The integrated model improved performance, achieving 87.5% accuracy while reduc-ing Type 1 errors to 0%. Since Type I errors – misclassifying unsound banks as sound – pose greater supervisory risk, minimizing them is crucial. The results confirm the su-periority of the logit model and demonstrate that integrating models enhances robust-ness and predictive reliability. The proposed integrated model can serve as a practical tool for supervisory and regulatory authorities to detect potential bank failures early and strengthen financial stability.