Graph-based feature selection and representation for multi-class classification tasks

Shir Friedman, M.Sc. student at the Department of Industrial Engineering

26 April 2022, 14:00 PM, Room 206& via zoom

Abstract:
Over recent years, feature selection has gained more attention, especially with the growing data complexity and volume. It has become a common processing step in high-dimensional datasets in order to find an optimal subset of relevant features in the data. There are a few approaches for feature selection, one of them is filter-based methods. In these algorithms, a score is calculated for all features and the selection is independent of the classifier. An example of such a score is the Jeffries-Matusita (JM) distance. It is a measure of separability between two distributions, previous work relay on this measure for filter-based feature selection. For each feature, JM calculates the separability strength between pairs of classes. A low-dimensional representation of the JM measures of the feature space is formed by diffusion maps, which is a non-linear dimension reduction technique. Diffusion maps organize the features by their class separation abilities. Moreover, feature elimination is performed based on the distribution of the points in the low-dimensional space. The algorithm was tested on 3 public datasets, and the results were compared to known filter-based feature selection techniques.

Bio:
Shir Friedman, M.Sc. student at the Department of Industrial Engineering in Tel Aviv University, specializing in Data Science. Shir holds a M.Sc in Management of Technology and Information Systems from Tel-Aviv University and a B.Sc. degree in Industrial Engineering from the Technion. Her research focuses on graph-based feature selection and representation. This research was supervised by Dr. Neta Rabin and collaborated with Dr. Lior Noy.