Business failure prediction based on two-stage selective ensemble with manifold learning algorithm and kernel-based fuzzy self-organizing map

摘要

More and more models and algorithms are used to predict business failure. Many of them are not suitable for the complicated distribution of financial data, which leads to unsatisfactory prediction performance. The manifold learning algorithm is a valid method to preprocess financial data because of the good performance on dimensionality reduction for any data distribution. The kernel-based method is introduced to improve the disadvantage of fuzzy self-organizing map (FSOM) which is the limitation of spherical data distribution. Therefore, this study adopts manifold learning algorithm to select feature subsets, and employs the kernel-based FSOM (KFSOM) to compose base classifiers, and proposes the two-stage selective ensemble model for business failure prediction (BFP). First, three manifold learning algorithms, which are Isomap, Laplacian Eigenmaps and Locally Linear Embedding, are adopted to select three feature subsets from original financial data. Then, KFSOM uses three kinds of kernel functions respectively, which are Gaussian, Polynomial and Sigmoid, to obtain three classifiers. Hence, three feature subsets are computed by KFSOMs with three kernel functions respectively to acquire nine base classifiers. Last, nine base classifiers are integrated by the two-stage selective ensemble method. In the first stage, nine base classifiers are ranked according to three standards. The stepwise forward selection approach is adopted to selectively integrate nine base classifiers according to different standards. In the second stage, three selective ensembles in the first stage are integrated again to acquire the final result. In the empirical research, this work employs financial data from Chinese listed companies to predict business failure, and makes comparative analysis with previous methods. It is the conclusion that the two-stage selective ensemble with manifold learning algorithm and KFSOM is good at predicting business failure.