Feature selection of Human Daily Activities using Ensemble method Classification
In Human activity recognition (HAR) research study it is a common practice using a wearable sensor to acquire the signal of human daily activities. In this study, database from smartphone inertial sensors is analysed for six different activities recognition. The aim of this paper is to compare diffe...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Conference or Workshop Item |
| Published: |
Institute of Electrical and Electronics Engineers Inc.
2019
|
| Online Access: | https://www.scopus.com/inward/record.uri?eid=2-s2.0-85075631958&doi=10.1109%2fSCORED.2019.8896253&partnerID=40&md5=52a5ddd56377dc4d42d24883f90f8516 http://eprints.utp.edu.my/24912/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In Human activity recognition (HAR) research study it is a common practice using a wearable sensor to acquire the signal of human daily activities. In this study, database from smartphone inertial sensors is analysed for six different activities recognition. The aim of this paper is to compare different filter method feature selections for multiclass problem based on human daily living activities using the smartphone inertial sensor. Three components for HAR processing stage are involved that comprises of data filtering and segmentation, data feature extraction, feature selection of the data and classification. An ensemble method of Random subspace with Support vector machine is adapted for classification. Model evaluation of holdout and 10-fold cross-validation methods are implemented for classification assessment. The performance of all human daily activities is evaluated according to comparison of overall accuracy for four type filter method feature selection method. From the result findings, the number of features that reduce to 198 feature archived 98.89 compared to 561 numbers of features 98.74 of overall accuracy for holdout method. While using the 10 cross validation method, the numbers of features are reduced to 424 with the overall accuracy 99.28 compared to 561 number of features with 99.22 of overall accuracy. © 2019 IEEE. |
|---|