Efficiency of gradient boosting decision trees technique in polish companies’ bankruptcy prediction

Abstract

The goal of the research was to compare the selected traditional bankruptcy prediction models, namely linear discriminant analysis and logit (logistic) models, with the technique called Gradient Boosting. In particular, the paper verifies two research hypotheses (verification was based on the balanced sample of Polish companies): [H1]: Gradient Boosted Decision Trees algorithm is more accurate than traditional bankruptcy prediction models: logit and discriminant analysis; [H2]: Boosted Decision Trees use both: financial ratios and normalized data from financial statements, but the same accuracy one can achieve only with the normalized data and the bigger number of weak learners.