Research On Genetic Algorithm-based Feature Selection For Prediction Of Enterprise Financial Crisis

With the rapid development of global integration, there are many internal and external environment of uncertain factors, leading to more and more crisis and challenges that enterprises fact. Therefore, enterprise financial crisis prediction is particularly important in a timely manner as the competition between enterprises is becoming fiercer. However, with the development of information technology, many industries including finance industry have produced and accumulated a large amount of data. A huge amount of data and high dimension of data have caused a lot of trouble for data processing, the reduction of data dimension in data processing from all walks of life appears to be increasingly important. Therefore, to find a feasible data feature selection method, choose the optimal feature combination, is of great benefit to the prediction of enterprise financial crisis.In this paper, the genetic algorithm is used for feature selection model, which includes a hybrid feature selection methods, namely first filter techniques and then wrapper techniques for the feature selection. In the filter selection techniques, this paper uses a double standards selection model, which consists of the information gain and the correlation coefficient, to get a preliminary feature subset after screening. And then the paper applies the genetic algorithm to combinatorial optimization in order to obtain the best feature combination for prediction effect. In the process of the genetic algorithm, using MATLAB as a modeling tool, BP artificial neural network is presented as the evaluation function of prediction accuracy. The improved hybrid feature selection model based on the genetic algorithm is established.In this paper, the financial data in 2014 of the domestic 304 enterprises from Shenzhen and Shanghai stock market is selected as the experimental data, which is used in the above model. The optimal feature subset and its forecast accuracy for the financial crisis prediction model of domestic listed companies are presented finally. Meanwhile, the efficiency of this model is confirmed. Afterwards, in order to verify the optimization of this model, the proposed feature selection model in this paper is used to compare with general feature selection model. Finally, the optimal feature subset is applying to analyzing the data of domestic listed companies in 2015 and predicting enterprise financial condition in 2017.The research in this paper presents a new feature selection model and the efficiency and optimization of this model have been confirmed. A new feature selection method can be used for the data processing of enterprise financial crisis prediction. The model of enterprise financial crisis prediction has been optimized during this paper, which will be more advantageous for companies to avoid crisis and for investors and creditors to predict and assess the development of the companies.