%0 Conference Paper %1 akter2018activity %A Akter, Syeda S. %A Holder, Lawrence B. %A Cook, Diane J. %B Internet of Things -- ICIOT 2018 %C Cham %D 2018 %E Georgakopoulos, Dimitrios %E Zhang, Liang-Jie %I Springer %K activity-recognition gps graph-mining graphical-features real sensor sensor-network smartphone %P 45-55 %R 10.1007/978-3-319-94370-1_4 %T Activity Recognition Using Graphical Features from Smart Phone Sensor %U https://link.springer.com/chapter/10.1007/978-3-319-94370-1_4 %X We develop a graphical feature-based framework that collects data from different kinds of sensor networks, represents the sensor network data as a graph, extracts graphical features from the graph representation, and adds those features to a set of non-graphical features that are typical for the application. Our hypothesis is that the addition of a structural representation and transitional features will improve performance for the corresponding prediction tasks of different networks. We apply our graphical feature-based approach on smart phone GPS sensor data to predict activities performed by phone users. We represent the location category corresponding to each GPS value as a node and movement of users from one GPS location to another as an edge in graph. Then we extract graphical features such as existence of nodes and edges from the graph representation and add them to basic sensor data features coming from the smart phone. We find that using this augmented feature set improves activity recognition accuracy by 7.27\% compared to using only basic non-graphical features with feature set augmented with existence of nodes performing the best. %@ 978-3-319-94370-1

@inproceedings{akter2018activity, abstract = {We develop a graphical feature-based framework that collects data from different kinds of sensor networks, represents the sensor network data as a graph, extracts graphical features from the graph representation, and adds those features to a set of non-graphical features that are typical for the application. Our hypothesis is that the addition of a structural representation and transitional features will improve performance for the corresponding prediction tasks of different networks. We apply our graphical feature-based approach on smart phone GPS sensor data to predict activities performed by phone users. We represent the location category corresponding to each GPS value as a node and movement of users from one GPS location to another as an edge in graph. Then we extract graphical features such as existence of nodes and edges from the graph representation and add them to basic sensor data features coming from the smart phone. We find that using this augmented feature set improves activity recognition accuracy by 7.27{\%} compared to using only basic non-graphical features with feature set augmented with existence of nodes performing the best.}, added-at = {2019-11-14T05:22:34.000+0100}, address = {Cham}, author = {Akter, Syeda S. and Holder, Lawrence B. and Cook, Diane J.}, biburl = {https://www.bibsonomy.org/bibtex/26355b3e69c41666628311c75b3b58dd0/jpmor}, booktitle = {Internet of Things -- ICIOT 2018}, day = 17, description = {Activity Recognition Using Graphical Features from Smart Phone Sensor | SpringerLink}, doi = {10.1007/978-3-319-94370-1_4}, editor = {Georgakopoulos, Dimitrios and Zhang, Liang-Jie}, eventtitle = {International Conference on Internet of Things}, interhash = {49a0c9fd1fb1949c47c095dbbd642dba}, intrahash = {6355b3e69c41666628311c75b3b58dd0}, isbn = {978-3-319-94370-1}, keywords = {activity-recognition gps graph-mining graphical-features real sensor sensor-network smartphone}, language = {English}, month = {06}, pages = {45-55}, publisher = {Springer}, timestamp = {2020-10-07T13:36:50.000+0200}, title = {Activity Recognition Using Graphical Features from Smart Phone Sensor}, url = {https://link.springer.com/chapter/10.1007/978-3-319-94370-1_4}, year = 2018 }