Human Activity Recognition in AAL Environments Using Random Projections

Table 10

Summary of HAR results using USC-HAD dataset.

Reference	Features	Classification method	Accuracy
Zheng [66]	Means and variances of magnitude and angles of acceleration along -, -  &  -axes	Least Squares Support Vector Machine (LS-SVM), Naïve-Bayes (NB)	95.6%
Sivakumar [67]	Accelerometer and gyroscope data	Symbolic approximation	84.3%
Vaka [68]	Mean, std. dev., correlation between   &  ,   &  , and   &  , and RMS	Random Forest	90.7%
This paper	99 times, frequency and physical features	Heuristic (random projections + PDFs + Jaccard distance)	95.52%