People in the Data Mining and Business Intelligence Lab at the Western University have been doing extensive research in the following areas:

  • Active learning of various types

Active Learning considers that the learning model could actively construct or select examples, so as to significantly speed up the learning process. Compared with the traditional supervised learning (i.e., passive learning in most cases), active learning may hold the key for solving the data scarcity problem (i.e., the lack to labeled data). More specifically, we propose a new active learning paradigm — active learning with generalized queries. In this learning paradigm, it is assumed that the oracles (often human experts) are capable of answering generalized queries (instead of only specific ones in the traditinal active learning). As one generalized query often represents a set of specific ones, active learning with generalized queries can further improve the learning effectively and quickly. See “Publications” for details.

  • Semi-supervised learning such as co-training

Semi-supervised learning tends to sufficiently utilize both labeled and unlabeled examples for building accurate learning models. Under various assumptions, semi-supervised learning algorithms attempt to “guess” the true labels for the unlabeled examples, and furthermore use them to augment the learning model. We specifically study when co-training works on real-world data in our research. See “Publications” for details.

  • Mining imbalanced, and cost-sensitive data
  • Human learning modeling
  • Machine learning for software engineering
  • Mining financial data, stock data, CRM (Customer Relationship Management)
  • Mining for Internet, and from webpage data
  • Learning Bayesian networks for optimal decision making
  • Evolutionary algorithms
  • Machine learning measures

QR Code
QR Code research (generated for current page)