Multi-hyperplane scoring model

Abstract

Multi-hyperplane scoring model is a decision model that classifies target data sets into a positive group and a negative group by minimizing the misclassification error using multiple hyperplanes. A collection of hyperplanes are used to divide a space into positive and negative regions. Each hyperplane is created in order to minimize the misclassification rate. Initially, all variables are ranked according to their entropies. The first half of variables with lower entropies is selected to generate the first pair of hyperplanes. The rest of variables are used during the subsequent stages. Our experiments compare the performance of a multi- hyperplane scoring model with a two-stage least cost credit scoring model. The result shows that our model has better accuracy than a two-stage least cost credit scoring model. Moreover, our model has comparable accuracy with decision tree, multilayer perceptron, linear discriminant analysis and support vector machine. However, we suggest using this model with small and medium size data sets since the construction of this model requires a long computation time.