Composite Classifiers for Bankruptcy Prediction

Composite Classifiers for Bankruptcy Prediction

Efstathios Kirkos (ATEI of Thessaloniki, Greece)
Copyright: © 2014 |Pages: 10
DOI: 10.4018/978-1-4666-5202-6.ch043

Chapter Preview



Recent developments in the field of intelligent bankruptcy prediction have pointed out that the combination of a number of techniques may yield models able to outperform individual classifiers. The combination of individual techniques is not a trivial task. Several methodological issues give rise to alternative approaches and may lead to different categorizations. A common categorization of composite classifiers divides them into ensemble and hybrid ones. In what follows, a brief introduction to these types of classifiers is presented.

According to the ensemble approach a number of different classifiers, each of which solves the same original problem, are trained. The individual decisions are aggregated and a final classification decision is reached.

Since it is pointless to multiply the same original classifier, the individual base classifiers must substantially differ. This diversification can be achieved in a number of ways.

  • Employment of different methods and development of corresponding models. Neural Networks, Decision Trees and Bayesian Networks are examples of individual methods. Normally, all models are trained by using the same samples.

  • Employment of different training sets. The idea is to create alternative data sets from an original data set and to train corresponding models. Bagging (Breiman, 1996) is a common example of this case.

  • Employment of different subsets of features. Different models are trained by using different input variables. This can be achieved by using different feature selection techniques.

  • Different initial settings of the base method. The same technique and the same data are used. The difference arises from the tuning of the base method. For a neural network for example this may mean different topologies, learning rates, training epochs etc.

Key Terms in this Chapter

Clustering: Form of data analysis that groups observations to clusters. Similar observations are grouped in the same cluster, whereas dissimilar observations are grouped in different clusters. As opposed to classification, there is not a class attribute and no predefined classes exist.

Hybrid Classifier: Classification technique that involves a usually small number of heterogeneous methods, which act complementarily to each other. Each method solves a different task and the classification decision is reached by one method.

Ensemble Classifier: Classification technique that aggregates the decisions of many individual classifiers. Each individual classifier solves the same original task and contributes its decision to the aggregation scheme.

Bankruptcy: The condition in which an organization fails to meet its financial obligations and thus must undergo debt reorganization or assets liquidation.

Classification: Form of data analysis that models the relationships between a number of variables and a target feature. The target feature contains nominal values that indicate the class to which each observation belongs.

Intelligent Bankruptcy Prediction: The employment of advanced statistical or artificial intelligence techniques for the prediction of bankruptcy cases.

Evolutionary Algorithms: Population-based optimization algorithms in which each member of the population represents a candidate solution. In an iterative process the population members evolve and are then evaluated by a fitness function. Genetic Algorithms and Particle Swarm Optimization are examples of evolutionary algorithms.

Complete Chapter List

Search this Book: