基于改进粒子群算法的模糊聚类-概率神经网络模型的企业财务危机预警模型研究

Use of Hybrid Fuzzy c-means and Probabilistic Neural Network Based on Improved Particle Swarm Optimization in the Prediction of Financial Distress

为了充分发挥概率神经网络在企业财务危机预警中的作用,克服概率神经网络平滑参数难以确定和空间复杂度高的不足,本文提出一类新的参数动态调整的粒子群算法优化概率神经网络的平滑参数,进而采用改进粒子群算法优化初始隶属度矩阵的模糊聚类方法实现对样本的选择,解决了概率神经网络平滑参数的确定及空间结构复杂的问题。提出了基于改进粒子群算法的模糊聚类-概率神经网络企业财务危机预警模型,并以我国上市公司作为研究对象进行了实证研究。结果表明,经过模糊聚类和改进粒子群算法优化的概率神经网络具有更优的预测性能,并在企业财务危机长期预警方面具有一定效用。

In order to investigate the role of probabilistic neural network （PNN） in the prediction of financial distress, and to overcome the difficulties caused by the smoothing parameter estimation and the high space complexity of the existing PNN, a novel adjustable parameter particle swarm optimization is proposed to optimize the smoothing parameter of PNN. Besides, an improved fuzzy c-means based on adjustable parameter particle swarm optimization is employed to achieve the instance selection for financial distress prediction, and then the combination of adjustable parameter particle swarm optimization and improved fuzzy c-means method are employed to help to overcome the shortcomings of PNN. The method hybridizing fuzzy -means and PNN based on improved PSO is proposed in the prediction of financial distress. An empirical study of listed companies in China is conducted, and the evaluation of the proposed model is validated. The results showed that the proposed method has a superior capacity in financial distress prediction compared with other artificial intelligent methods, such as neural network, decision tree, and support vector machine. In addition, the proposed method also improves the long- term financial distress prediction performance.