Human Activity Dataset Generation - Smartphone-Based Human Activity Recognition - page 73

4.5 Summary
In this chapter, a new dataset for HAR using smartphones has been introduced. We
thoroughly described the process to achieve this by incorporating the most important
stages such as device selection, data collection experiments, dealing with data and
validation. We also acknowledged initial classification results using 6 up-to-date ML
algorithms including a multiclass SVM. This latter approach showed a noticeable
advantage in terms of classification accuracy confirming our purpose of using it as
the core ML algorithm in this thesis. This was also confirmed by the organized HAR
competition whose winning algorithm was SVM-based.
These findings allows to argue that the use of smartphones for motion information
retrieval seems feasible. In addition, it is also less obtrusive and invasive than other
special purpose solutions (e.g. wearable sensors), and a practical way to walk for
effectively performing HAR.
Making the data available to the public has brought many advantages. First, it
offers to the research community the opportunity of comparing different HAR related
works based on the same data, therefore, ML methods can be better evaluated. Sec-
ond, it provides feedback from many users regarding different aspects of the dataset
such as signal processing, feature selection, and possible corrections in future ver-
sions of the dataset.
References
C.B. Abdelkader, R. Cutler, L. Davis, in Stride and Cadence as a Biometric in Automatic Person
Identification and Verification , International Conference on Automatic Face and Gesture Recog-
nition, 2002
Android (2013) Android developers. http://developer.android.com/index.html . Accessed 05 Nov
2013
M. Aupetit, Nearly homogeneous multi-partitioning with a deterministic generator. Neuro Comput
72 , 1379-1389 (2009)
K. Bache, M. Lichman, UCI machine learning repository (2013) http://archive.ics.uci.edu/ml
L. Bao, S.S. Intille, Activity recognition from user-annotated acceleration data. Pervasive. Comput.
(2004)
B. Bruno, F. Mastrogiovanni, A. Sgorbissa, T. Vernazza, R. Zaccaria. in Human Motion Modelling
and Recognition: A Computational Approach , IEEE International Conference on Automation
Science and Engineering, 2012
C.C. Chang, C.J. Lin, LIBSVM: A Libraries for support vector machine. ACM Trans. Intell. Syst.
Technol. 2 , 27-54 (2011)
S. Dernbach, B. Das, N.C. Krishnan, B.L. Thomas, D.J. Cook, in Simple and Complex Activity
Recognition Through Smart Phones , International Conference on Intelligent Environments, 2012
P. Duhamel, M. Vetterli, Fast fourier transforms: a tutorial review and a state of the art. Sig. proce.
19 , 259-299 (1990)
A. Gupta, C. Milanesi, R. Cozza, C.K. Lu, Market share analysis: mobile phones, worldwide, 2q13.
Technical report, Gartner Inc. (2013)
M. Hall, E. Frank, G. Holmes, B. Pfahringer, P. Reutemann, I.H. Witten, The weka data mining
software: an update. ACM SIGKDD Explor. Newslett. 11, 10-18 (2009)