Ensembles of Local Linear Models for Bankruptcy Analysis and Prediction

Abstract

Bankruptcy prediction is an extensively researched topic. Also ensemble methodology has been applied to it. However, the interpretability of the results, so often important in practical applications, has not been emphasized. This paper builds ensembles of locally linear models using a forward variable selection technique. The method applied to four datasets provides information about the importance of the variables, thus offering interpretation possibilities. Bankruptcy prediction has gained increasing interest since the 1960s [1], and not without reason. Predicting the financial distress of firms benefits the company leaders by identifying internal problems, but also assists auditors in their work for finding potentially troubled firms. Above all, bankruptcy prediction produces information for investors and banks so that they can make sounder lending and