A Comparison of Feature Extraction Methods for the Classification of Dynamic Activities From Accelerometer Data

Abstract

Abstract-Driven by the demands on healthcare resulting from the shift toward more sedentary lifestyles, considerable effort has been devoted to the monitoring and classification of human activity. In previous studies, various classification schemes and feature extraction methods have been used to identify different activities from a range of different datasets. In this paper, we present a comparison of 14 methods to extract classification features from accelerometer signals. These are based on the wavelet transform and other well-known time-and frequency-domain signal characteristics. To allow an objective comparison between the different features, we used two datasets of activities collected from 20 subjects. The first set comprised three commonly used activities, namely, level walking, stair ascent, and stair descent, and the second a total of eight activities. Furthermore, we compared the classification accuracy for each feature set across different combinations of three different accelerometer placements. The classification analysis has been performed with robust subject-based cross-validation methods using a nearest-neighbor classifier. The findings show that, although the wavelet transform approach can be used to characterize nonstationary signals, it does not perform as accurately as frequencybased features when classifying dynamic activities performed by healthy subjects. Overall, the best feature sets achieved over 95% intersubject classification accuracy.

Optimal Recognition Method of Human Activities Using Artificial Neural Networks

“…For feature extraction, the most used methods may be on one hand statistical methods (max, media, variance, standard deviation, energy), called also time-domain features, and on the other hand frequencydomain features using FFT or time-frequency features using wavelet filter banks. A comparison between these methods is presented in [4]- [6].…”

“…Other classification methods encountered are Support Vector Machines (SVM), Decision Trees classifiers, knearest neighbor (kNN) classifier [4], hidden Markov models, and neural networks [6]. A comparison between classification methods is presented in [9] and [10].…”

Feature Analysis to Human Activity Recognition

“…For instance, Preece at al. used 2 seconds long window with 50% overlapping while Gao et al and Karantonis et al used 1 second long window without overlapping [4,8,14]. In our case, a window contains 32 samples and there is a 50% overlapping between windows in the training phase and no overlapping in the test phase.…”

“…During the data acquisition, 5 sensor nodes were placed at multiple body locations of each person: left and right forearm, waist, left and right ankle. In this study we used the data of only one sensor which has been placed on the right ankle because Ertugrul et al and Oniga et al demonstrated that one sensor is enough for appropriate HAR recognition while Preece et al claimed that the ankle is an optimal placement for single sensor [8,28,31].…”

“…In our case, a window contains 32 samples and there is a 50% overlapping between windows in the training phase and no overlapping in the test phase. This size covers 1.6 seconds time interval because the sampling frequency of the [8]. At the rest of this subsection we give a short description to the methods in Table 1.…”