Classification algorithms for modeling economic choice

The article shows how some novel machine learning algorithms can be applied to economic problems of discrete binary choice. An examination is made of three typical business tasks – classifying overdraft applications, credit risk management, and marketing segmentation. Both traditional econometric methods (logistic regression and linear discriminant analysis) as well as five more advanced machine learning algorithms (neural networks, k-nearest neighbours, naive Bayes classifier, random forest, and support vector machine) have been used for modelling these tasks. For all the classification tasks, the random forest algorithm robustly registers improved forecasting accuracy over the more traditional approaches. This underlines the need to supplement the classical econometric toolbox with innovative methods, with the random forest, the support vector machine, and the neural network being prime candidates.