Mining high-dimensional business data is a challenging problem. Particularly in bankruptcy predictions, we need to analyze large amounts of information from financial statements and stock markets. This paper proposes a new strategy to deal with the problem. Because of the highly correlation among financial information, this study employed a technique called generalized discriminant analysis (GDA) to identify important features and reduce the data dimension. GDA is a nonlinear discriminant analysis using kernel function operator. It’s easy to deal with a wide class of nonlinearity in financial data, and can reduce the computational loading of subsequent prediction classifier. Due to the promising success of kernel machines in many applications, this study utilized a generalized multiple kernel machine (GMKM) to serve as the predictor. Combining the strengths of GDA and GMKM, our system robustly outperforms traditional prediction systems.
All Science Journal Classification (ASJC) codes
- Business and International Management