Comparison of wrapper and filtering approaches for corporate failure prediction

Outbreak of debt crisis in Europe has made the issue of corporate failure prediction, known as financial distress prediction (FDP) as well, a significant topic in the field of management science. The purpose of this paper is to propose five hybrid classifiers to tackle corporate failure prediction problem. Principle component analysis (PCA),information gain (IG) and relief (Re) methods as representatives of feature selection filtering approach, and genetic algorithm (GA) and particle swarm optimization (PSO) techniques as the representatives of feature selection wrapper approach, have been integrated with k-nearest neighborhood (k-NN) to create our five classifies for our given data set. According to results, PSO-kNN ensemble classifier outperformed all the applied classifiers in the literature in terms of prediction accuracy for our defined data set.