Human Activity Recognition on Smartphones using a Multiclass Hardware-Friendly SVM (Part 1)
This blog is divided into two parts, In this part we will attempt to summarize all the concepts one-by-one from Human Activity Recognition on Smartphones using a Multiclass Hardware-Friendly SVM research paper. And in the second blog we will try to solve Human activity recognition problem based on this research papers approach.
Link to the research paper 👉 Click Here
Abstract
To get Human activity readings, sensors were attached to the human bodies(subjects). Smartphones are inertial sensors but they are limited in terms of energy and computational power so David Anguita and his team proposed a novel hardware-friendly approach for multiclass classification. This method adapts the standard SVM and exploits fixed-point arithmetic for computational cost reduction. It gives a significant improvement in terms of computational costs while maintaining similar accuracy.
Summary
ML methods that have been previously employed for recognition includes Naive Bayes, SVMs, Threshold-based and Markov chains.
To Understand Multiclass hardware-Friendly SVM concept, we will first understand What is Hardware-Friendly SVM (HF-SVM)? and What is Multiclass?*
Hardware-friendly SVM (HF-SVM) concept was first introduced by Angita et al. This method exploits fixed-point arithmetic in the feed-forward phase of the SVM classifier, so as to allow the use of this algorithm in hardware-limited devices.
In Multiclass they have chosen One-Vs-All (OVA) method because it's accuracy is comparable to other classification methods and its learned model uses less memory when compared for instance to One-Vs-All (OVO) method.
Experimental Results
To evaluate the performance of the algorithm, a set of experiments were carried out using the AR dataset, which was collect from the subjects. The results are in the following image,
The experiment shows that for this dataset k = 6 bits are sufficient for achieving a performance comparable with the MC-SVM approach that uses 64- bit floating-point arithmetic.
The results of confusion matrix between MC-SVM and MC-HF-SVM are as following for k = 8,
Conclusion
The MC-HF-SVM method is an appealing approach for use in Ambient Intelligence systems for healthcare applications such as activity monitoring on smart-phones.