Seminar - Evolutionary Computation for Feature Manipulation in Classification on High-dimensional Data

ECS PhD Proposal

Speaker: Binh Ngan Tran
Time: Tuesday 27th January 2015 at 09:00 AM - 10:00 AM
Location: Cotton Club, Cotton 350

Add to Calendar Add to your calendar

Abstract

Classification problems are typically presented with a large set of features and the class labels. This feature set may contain redundant and irrelevant features to the target concepts. Feature manipulation is the process of selecting relevant features or constructing high-level features from the original ones to improve the quality and performance of a classification algorithm. Although feature manipulation has been studied for decades, the task on high-dimensional data with thousands to tens of thousands of features is still challenging due to the fact that the existing methods usually face the problem of being stagnation in local optima and consume much computational time. Evolutionary computation (EC) techniques are well-known for their global search. To be specific, Particle swarm optimisation (PSO) has been successfully applied to feature selection and Genetic programming (GP) has shown its ability of constructing better discriminating features. However, applications of these EC techniques on high-dimensional data require high memory and much computational time. Therefore, this research aims to develop a PSO based approach to feature selection and a GP based approach to feature construction in classification on high-dimensional data to further enhance the performance of feature manipulation. A combination of these two techniques will also be investigated.

Go backGo back to the seminar list