Balance sheet and income statement provide potentially vast volumes of information. Despite the large number of predictive variables, in most cases, user cannot make a judgment easily about the survival of a company. In this paper, we indicate a set of useful variables for failure prediction by stepwise discriminant analysis (SDA). Furthermore, this study applied a data mining technique to explore and compare the performance of Particle Swarm Optimization (PSO), Classification and Regression Tree (CART) and Support Vector Data Description (SVDD). Results show that PSO does not significantly differ from CART, but due to lower average error rate, PSO is more efficient than CART and PSO and CART significantly perform SVDD.

Keywords