Feature Grouping Based On Ga And L-Gem For Human Activity Recognition

Human Activity Recognition is useful in many applications such as video surveillance and health care for elderly. Multiple Classifier System (MCS) has been proved to be better than single classifiers theoretically and empirically. Diversity is the key to improving the performance of an MCS. Feature...

Full description

Saved in:
Bibliographic Details
Main Authors: Xue, Yi-Wen, Liu, Jing, Chen, Jiamin, Zhang, Yun-Tao, Cao, Renhua
Format: Conference Proceeding
Language:English
Subjects:
Activity recognition
Classification algorithms
Genetic algorithm
Genetic algorithms
Human activity recognition
Localized generalization error model
Multiple classifier system
Robot sensing systems
Search problems
Training
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Human Activity Recognition is useful in many applications such as video surveillance and health care for elderly. Multiple Classifier System (MCS) has been proved to be better than single classifiers theoretically and empirically. Diversity is the key to improving the performance of an MCS. Feature grouping is one of the common methods. In this paper, we propose a new feature grouping method that uses Localized Generalization Error Model(L-GEM) as an evaluation criterion for optimizing MCS. Genetic Algorithm (GA) is used to evolve the weights of the members of MCS so that a set of diverse classifiers is combined to build an MCS which minimizes the localized generalization error. Experiments is performed based on a human activity recognition dataset and the results are compared with random subspace method to show that the MCS performs better.
ISSN:2160-1348
DOI:10.1109/ICMLC.2018.8527017