Evaluation of Quality of Neural Network Models and Discriminant Analysis in ROPO Forecasting

Abstract

In the paper the deep learning model based on deep neural networks for predicting the ROPO behaviour of tourists is compared with classical discriminant analysis techniques as: linear discriminant analysis, kernel discriminant analysis, KNN method, SVM and classification trees on a real dataset containing the results of survey prepared by authors. In the second part of the paper, the methods of deep neural network tunning will be used on the same dataset and their effectiveness in improving the quality of the model will be assessed. The achieved results will show that there are some situations when deep learning gives better results than classical methods of discriminant analysis for economic datasets (like the dataset describing the ROPO behaviour in this study) but it requires additional research on why certain combinations of net architecture, loss function, optimizers and selected parameters behaves better than others.