In recent years, the global economic recession, followed by failure of business due to poor management, has resulted in a domino effect that occurred throughout the financial system. To avoid the expansion of loss, the issue of business failures should be seriously considered. In this paper, firstly, to reduce the time and space spent in our models learning and prediction, we use the data mining methods -stepwise regression, genetic algorithms and self-organizing map network for pre-processing data. Secondly, to match with the food searching behavior of the fruit fly, we modify Pan’s optimization algorithm to a three-dimension space for the General Regression Neural Network (FOAGRNN), then, compared it with the Backpropagation Neural Network, Genetic Programming, General Regression Neural Network (GRNN),and the traditional Least Square method for financial distress forecasting models. Finally, through a substantial number of experiments, we realized that if we wanted to study the company's financial crisis early warning, in addition to considering the company's financial variables, corporate governance variables should not be neglected. Besides, we also found that our modified 3D-FOAGRNN outperformed the General Regression Neural Network, Genetic Programming, Backpropagation Neural Network and the Least Square method in terms of forecasting accuracy.