Engineering HDR Seminar Series 24_2013: A New Multivariate Alternating Decision Tree Learning Algorithm

Mr Sok Hong Kuan, ECSE postgraduate student

Date: 2013-07-24
Time: 11:00 to 12:00
Venue: Engineering Meeting Room 1, 5-4-22


Alternating decision tree (ADTree) is a classification model which consists of a set of decision rules. These decision rules form a special tree representation that inherently supports boosting, a successful and important development in the field of machine learning for majority voted prediction. ADTree is graphically comprehensible and interpretable like those classical decision trees, C4.5 and CART despite having a slightly different tree structure. There are many research works on different aspects of decision trees in the literature but some of these important studies are missing in the ADTree literature. Existing researches on ADTree use only univariate decision nodes where each of these is based on a single feature. This research aims to provide a complementary study using multivariate decision nodes for ADTree. The next study is on sparse representation of multivariate ADTree. Sparseness criterion is to be imposed on multivariate decision nodes using regularization technique known as elastic net for automatic feature selection. The last study is on generating omnivariate ADTree where the decision node can be univariate or multivariate. The idea is to match the decision nodes to the complexity of the sub-problems formed through tree partitioning instead of assuming the same inductive bias for all decision nodes. A suitable model selection technique for ADTree will be studied in order to select the right type of decision node.

About the Speaker

Sok Hong Kuan received his BEng (Hons) degree with first class honors in Electrical and Computer Systems engineering from Monash University, Malaysia in 2010. He is currently pursuing his PhD with Monash University under the supervision of Dr Melanie Ooi Po-Leen and Dr Kuang Ye Chow. His research focus is on developing new multivariate ADTree algorithm, a machine learning algorithm that combines decision tree and boosting methodology. His main research areas include machine learning and classification.