A Simple Neural Network Model of University Preferences: Two Algorithms and a Case Study

Abstract

In this paper, we make use of two algorithms/methods, namely the neural networks and the multi-objective evolutionary fuzzy classifier to develop a simple model of university preferences. We use technology, teaching quality, research productivity, managerial quality, physical capital and social capital as the input variables. We first construct a neural network with a hidden layer and determine the degrees of importance of the input variables in relation to the university preferences having dichotomous values signifying the positive and negative attitudes towards the relevant higher educational institutions. With the setup and the data, it turns out the teaching quality is the most influential factor followed by the managerial quality. The second algorithm we make use of is the multi-objective evolutionary fuzzy classifier. We choose 90% of the data for training and the rest (10%) testing purposes. We obtain a 93.33% accuracy, which is quite high. In sum, machine learning algorithms turn out to be fairly successful in modeling university preferences. The classification performance of the algorithms is remarkable. In an extended framework, we can reasonably expect that the forecasting models based on machine learning algorithms would also yield high degrees of accuracy. In addition to the algorithms exemplified in this paper, algorithms such as support vector machines, random forest and bagging are likely to produce results that could be of practical significance for managerial policy makers.