Fuzzy kernel k-medoids application with fisher score feature selection for predicting bank financial failure

Abstract

The bank financial failure has a huge impact on the real sector, households and can even cause knock-on effects for other banks, therefore is important to predict bank financial failure. Prediction bank financial failure is like an early warning system for bank, because a bankruptcy doesn't happen suddenly but there are indications that can be detected that is financial statement. Financial statement will be extracted to 6 components of CAMELS. The 2019, we had predicted bank financial failures using random forest with the data that used in Boyacioglu, Kara and Baykan paper (2009), however in this research we will make novelty by using fuzzy kernel k-medoids. Based on our results, fuzzy kernel k-medoids using RBF kernel with s = 0.1 and 60% composition of training data has 100% for accuracy, sensitivity, precision, specificity, and f-score with 0.9 sec running time. If we compare to our previous research by random forest, fuzzy kernel k-medoids gives the highest accuracy prediction, but if we compare to Boyacioglu, Kara and Baykan research (2009), it's has the same accuracy but with fuzzy kernel k-medoids, we can use only 60% of training data to learn.