A Multi-Label Classification With Hybrid Label-Based Meta-Learning Method in Internet of Things

Sung-Chiang Lin,Tsung-Ju Lee,Chih-Jou Chen

doi:10.1109/access.2020.2976851

Abstract

With the widespread adoption of Internet connected devices and the application of Internet of Things (IoT), more and more research efforts focusing on using machine learning techniques in recognizing activities from IoT sensors, especially in solving multi-label classification problems. Without considering the associations among labels, traditional approaches aim to transform the original multi-label classification problem into several single-label classification problems. The loss of information among labels will damage the classification performance. In this paper, we proposed a novel hybrid label-based meta-learning algorithm for multi-label classification based on an ensemble of a cluster algorithm and generalized linear mixed model (GLMM). In this algorithm, the clustering phase is performed to catch the association among labels and to reduce the computational complexity from vast label subsets simultaneously, and the GLMM phase is performed to solve dependence of a subject with multi-labels in training data. The numerical results show that the proposed algorithm outperforms others, especially for cases with relatively large number of labels.

Highlights

The Internet of Things (IoT) is a link and exchange network of messages formed by physical objects, such as vehicles, machines, household appliances, etc
STRATEGIES FOR MULTI-LABEL CLASSIFICATION In this paper, we focus on the second type, problem transformation methods, and propose a novel hybrid label-based learning algorithm for multi-label classification, which is based on an ensemble of a clustering algorithm and generalized linear mixed model (GLMM) [22]
We proposed a hybrid label-based learning algorithm which assembles a clustering algorithm and a generalized linear mixed model to deal with multi-label classification problems

Summary

INTRODUCTION

The Internet of Things (IoT) is a link and exchange network of messages formed by physical objects, such as vehicles, machines, household appliances, etc. B. STRATEGIES FOR MULTI-LABEL CLASSIFICATION In this paper, we focus on the second type, problem transformation methods, and propose a novel hybrid label-based learning algorithm for multi-label classification, which is based on an ensemble of a clustering algorithm and generalized linear mixed model (GLMM) [22]. STRATEGIES FOR MULTI-LABEL CLASSIFICATION In this paper, we focus on the second type, problem transformation methods, and propose a novel hybrid label-based learning algorithm for multi-label classification, which is based on an ensemble of a clustering algorithm and generalized linear mixed model (GLMM) [22] This approach overcome some limitations of existing label-based methods, since the association among labels and dependence of a subject with more than one labels in training data are considered and solved through clustering algorithm and GLMM.

THE PROPOSED METHOD

BETWEEN-GROUP META-LEARNING WITH GLMM

WITHIN-GROUP META-LEARNING WITH GLMM

RESULTS

DISCUSSION AND CONCLUSION