TY  - CONF
A1  - Nurhanim, K.
A1  - Elamvazuthi, I.
A1  - Izhar, L.I.
UR  - https://www.scopus.com/inward/record.uri?eid=2-s2.0-85080028278&doi=10.1109%2fROMA46407.2018.8986719&partnerID=40&md5=cb58f860a89991404dfc7f268ad4bcb7
PB  - Institute of Electrical and Electronics Engineers Inc.
SN  - 9781728103747
Y1  - 2018///
ID  - scholars9504
TI  - Ensemble Methods for Classifying of Human Activity Recognition
KW  - Accelerometers; Adaptive boosting; Animation; Computer games; Decision trees; Extraction; Feature extraction; Frequency domain analysis; Manufacture; mHealth; Random forests; Robotics; Signal processing; Smartphones; Time domain analysis; Walking aids; Wearable sensors
KW  -  10-fold cross-validation; Feature extraction and classification; Frequency domains; Human activity recognition; Nested dichotomies; Overall accuracies; Random subspaces; Rotation forests
KW  -  Classification (of information)
N1  - cited By 2; Conference of 4th IEEE International Symposium in Robotics and Manufacturing Automation, ROMA 2018 ; Conference Date: 10 December 2018 Through 12 December 2018; Conference Code:157694
N2  - Wearable sensors of a smartphone are often used in Human activity recognition (HAR) application. Recently, utilization of wearable sensors has raised more importance in several applications such as computer games, medical care animation and film making. In this paper, classification of HAR using different type of Ensemble methods has acquired database from 30 subjects using smartphone to perform six different daily activities. The signal processing stages involved filtering and segmentation, time domain and frequency domain of feature extraction and classification for walking, walking upstairs, walking downstairs, sitting, standing and laying. The classification involved ensemble methods of Bagging, Adaboost, Rotation forest, Ensembles of nested dichotomies (END) and Random subspace with Random Forest(RF) as base classifier. Holdout method and 10 fold cross validation method are executed to evaluate the classification of each activity. Each activity assessment was based on precision, recall, F-measure and overall accuracy. From the results of classification of all activities, it was found that Adaboost-RF classifier given overall accuracy rate 98.07 for holdout method and 98.82 10 fold cross validation method. © 2018 IEEE.
AV  - none
ER  -