Séance Séminaire

Séminaire des Doctorant·e·s

Support vector Machine approch and failure companies prediction

We analyze the financial determinants of firm failure and the classification accuracy of the prediction models. The aim of our research is to construct preventive alert system rather than a decision-making one. The theoretical and empirical literature suggests various conceptions tools for the failure prediction such as, Discriminate Analysis, Logistic Regression, Artificial Neural Network, Genetic Algorithms... As part of this research, we test the predictive accuracy of the SVM model regarding failed companies. The grid-search technique, using the cross-validation, is tested to find out the best parameter value of SVM’s kernel function. Our main empirical finding is that the SVM model gives high prediction accuracy relatively to the other tested predictive models.