Abstract: In order to forecast the corporate finance performance,we must choose the appropriate forecast method.The forecast model used widely lacks generalization ability.In this paper,we propose a modified version of support vector machines (called AdaBoost support vector machine) to forecast financial perform ance of Chinese listed companies.In the choice of kernel function of support vector machine,forecast re sults are measured for each kernel function and its associated parameters with a view of identifying the most appropriate forecasting model.The experiment results show that our AdaBoost-support vector ma chine model with rbf kernel function compares favorably to probabilistic neural network and decision tree model.We also construct sub-industry financial performance prediction model for different industry.We find that the test accuracy of different industry varies and estimating separate models for each industry do not result in models with a higher predictive accuracy than the global model.

Key words: financial performance, support vector machine, adaBoost algorithms