Influence Prediction Using Bayesian Belief Networks For Real Life Networks

Influential node identification could soon become a key practice to define the flow of information throughout any network. It has already been frequently attempted using graph theory approaches and optimization techniques approaches such as node centrality measures, label propagation, Naive-Bayes approaches and a lot more. In this paper, we try to use the generalised Bayesian Belief Networks to classify data from an open source data repository using the basic probability distribution of multiple variables. The intent is to accurately predict influential repositories from a collection of repositories after data cleaning, data processing and modeled semi-supervised learning. We also use Naive Bayes algorithm and and Simple Logistic Regression on the same data and found that the underlying mask of Belief functions in Bayesian Belief Networks predicts the influence of a given new data more accurately than the other two algorithms.