Table 6 Comparison with the reported activity recognition methods
From: Identifying typical physical activity on smartphone with varying positions and orientations
Reference | Smartphone position | Activities numbers | Contributions | Algorithm and accuracy | Limitations |
|---|---|---|---|---|---|
Anjum et al. [15], 2013 | Pant pocket, hand, hand bag, shirt pocket | 7 | Activity recognition with smartphone at multiple positions including pant pocket, hand, hand bag and shirt pocket | Decision tree (AUC 0.985) | Limited activity traces and thus would tradeoff the performance in external verification |
Arif [16], 2014 | Leg front pants pocket | 6 | Demonstration of better activity classification accuracy | 10-fold KNN (98.2%) | Position is fixed in front pants leg pockets |
Romain Guidoux et al. [17], 2013 | Leg front pants pocket | 9 | Estimation of total energy expenditure with phone-position independent by transform | Total energy expenditure (73.6%) | Low accuracy |
Yongjin Kwon et al. [19], 2014 | Pants pocket | 5 | Unsupervised learning without labels | Hierarchical clustering or DBSCAN (above 90% accuracy) | Some important activities including going upstairs and downstairs were not studied |
Sourav Bhattacharya [20], 2014 | Jacket pockets, pants pockets, backpack | 8 | Deal with unlabeled data | Sparse coding (80%) | Important activities including going upstairs and downstairs were not studied |
This paper | Any pockets | 5 | Automatically identify the locations of the smartphone and conveniently activity recognition with smartphone at any pockets | 10-fold J48 (89.6%) | More situations, such as in the hand, should be further studied |