Bayesian Network Classifiers

Abstract

The Naive Bayes classifier is probably the first classification algorithm students are taught to code. Richer network structures yield better results, for instance the limited dependence Bayesian classifiers of Sahami (1998). Interestingly, pursuing a discriminative task (classification) makes the building of Bayesian networks a lot simpler. As with most machine learning and statistics, the over-fitting problem is an issue for these. We have recently developed superior hierarchical smoothing methods for Bayesian network classifiers, as well as ensembling methods. Note the most popular classification algorithms currently in competitions like Kaggle are the Random Forest and Gradient Boosting (of trees), both being ensembling algorithms. Our new methods beat these algorithms handily, and are also able to scale quite simply. I will describe our new methods and their experimental evaluation. This is joint work with Francois Petitjean and PhD student He Zhang.

Biography

Wray Buntine is a full professor at Monash University from 2014 and is director of the Master of Data Science, the Faculty of IT's newest and in-demand degree. He was previously at NICTA Canberra, Helsinki Institute for Information Technology where he ran a semantic search project, NASA Ames Research Center, University of California, Berkeley, and Google, as well as several startups. He is known for his theoretical and applied work and in probabilistic methods for document and text analysis, social networks, data mining and machine learning. More details can be found at his website: https://topicmodels.org/about/