The influence of learning algorithms for Bayesian Networks on predictions

Abstract

In this thesis, attention is paid to building different Bayesian networks. You can think of aspects such as parameter learning, search procedures and score functions. In addition, a distinction is made between the use of Discrete Bayesian Networks and Gaussian Networks. These models both have different assumptions which are also discussed. Finally, the theory is applied to publication and citation data for a group of Canadian researchers. We will build Bayesian networks with different techniques and try to predict and compare the performance of researchers. We will also build an algorithm based on clustering that can perform predictions by using one of the possible learning algorithms.