Learning Algorithm of LS-SVM Based on Quadratic Renyi-Entropy and Empirical Analysis

This paper applies quadratic Renyi entropy to enterprise financial distress prediction and puts forward a learning algorithm of least squares support vector machines (LS-SVM) based on quadratic Renyi entropy. By respectively analysis and comparison of the algorithm with the traditional LS-SVM, the standard SVM, MLR and BP-ANN, we can see that this algorithm is significantly superior to other algorithms and has good stability, whether in the aspect of the number of training samples or the operation time. Empirical analysis shows that it is successful to apply quadratic Renyi entropy to enterprise financial distress prediction fields. At the same time, through doing test of significance and making factor analysis to initial input variable, the number of input variables decreases, but the prediction accuracy rate reaches 88%. So the method of factor analysis proves to be effective in this paper.