A Heterogenous Online Ensemble Classifier for Bankruptcy Prediction

Considering current economic and market conditions, an ensemble model that generates precise predictions of an organization's possible bankruptcy state is of utmost importance for strategic decision-making. This research paper aims to investigate and apply reputable ensemble classifiers to existing credit datasets to address our identified research problem, which states, 'existing ensemble-based bankruptcy forecasting model lack diversity. If classifiers from different base algorithms are combined, we can formulate the much-needed variety to produce the anticipated result of ensemble classifiers when it comes to attaining diversity'. This research paper invokes the use of three popular classification techniques, support vector machines, decision trees, and artificial neural networks; these techniques are combined using the bagging and boosting combination method. A varying number of combined classifiers are applied to each classifier ensemble to investigate if diversity plays a role in prediction performance. The results of this report have indicated that diversity plays a role in forming an accurate prediction model. Decisions tree and artificial neural networks ensembles that are applied with 100 and 20 classifiers, respectively, demonstrated increased accuracy rates and prediction performance. Therefore, the results of this research paper generate novel insights into the relationship between diversity and prediction accuracy.