Nonlinear fuzzy robust PCA algorithms and similarity classifier in bankruptcy analysis

In this article the problem of the bankruptcy analysis has been tackled using Artificial Intelligence based methods to two different data set; Australian credit card screening data set and Japanese credit card screening data set. Bankruptcy analysis is carried out so, that data is first preprocessed using principal component analysis algorithms and the resulting principal components are used as new data in the similarity classifier to differentiate whether credit card application should be accepted or is there a real danger of applicants bankruptcy and hence the credit card should not be granted. In this article, two, new nonlinear fuzzy robust principal component analysis algorithms (NFRPCA1 and NFRPCA2) are derived and their performances are compared to bankruptcy analysis using the above mentioned combination. Higher classification accuracy was received, with Australian credit data 88.39% and with Japanese credit data 86.59% than without these new methods. Moreover, with NFRPCA1 algorithm we could also get remarkable dimension reduction this way and hence reducing computational time required. This property can also be very useful with large high dimensional data sets.