In collaboration with: Leen-Kiat Soh
Meet with Soh and me before submitting a proposal on this
project
Online interviews yield audit trails that have numerous attributes per session or record. These attributes are often incomplete, noisy, and correlated. In order to better analyze interviews that lead to better interview outcomes, it is thus important to extract important features or feature sets before, for example, running a decision tree analysis on the data. This project seeks to investigate association rule mining as an approach to feature extraction and compares its effectiveness to conventional machine learning algorithms such as C4.5. Specifically, features or feature sets that yield strong rules (high support and high confidence) likely indicate that the features or feature sets are important. Further, rule mining is also able to discretize features into "meaningful" intervals creating sub-range of values paired with their features. This discretization, while increasing the number of feature combinations, facilitates homing in on certain intervals. Identifying an effective discretization scheme and analyzing its impacts are also vital in this project.
Return to the CSCE 478/878 Home Page