Classification of human activity based on smartphone inertial sensor using support vector machine

Ensemble One-vs-One SVM Classifier for Smartphone Accelerometer Activity Recognition

“…The following abbreviated labels have been used in tables III and IV: LA-Laying, SI-Sitting, ST-Standing, WA-Walking, UP-Walking upstairs, DO-Walking downstairs, ACT -Activity, PRE -Precision and ACC -Accuracy. The comparison of the correct accuracy of classification rate between different methods reported previously in literature is presented in table V. To reiterate, even though the one-versus-all polynomial approach presented in [8] performs slightly better than the proposed (H score and Neural Network), the implementation in [8] is not hardware friendly and will require more effort to parallelise the classifier. This is mainly because of the use of polynomial kernel in [8] which requires a number of floating-point multiplications and would consume significant amount of FPGA resources when mapped onto such a parallel platform.…”

“…For unbiased comparison, the UCI open repository has bee used to test our approach and compared with six other classification results presented in [8][20] [22] from 2012 to 2017. The dataset was randomly split into a training set (of 75%) and a validation set (of 25%).…”

“…Nurhanim et al [8] used the HAR dataset [20] to compare the performance of different kernel of classification for support vector machine. Two approaches were presented, the multi-class support vector machine polynomial kernel and multi-class support vector machine Linear Kernel using "winner take all".…”

“…Method Accuracy(%) [22] k−NN 90.63 [22] OVA SVM 93.72 [22] OVO SVM Linear 96.40 [20] OVA SVM Gaussian 96.50 Proposed High Scoring (Hscore) 96.90 [8] OVA Linear 96.91 Proposed Neural Network (NN) 97.30 [8] OVA Polynomial 98.57…”

“…Human activities have been classified into three main types in [8] as short event, basic and complex activities. To classify such activities, many methods have already been used [9], like the generic architecture for big data healthcare analytic [10], Map Reduction [11], Nave Bayes [12] and Machine Learning (ML) [13].…”