*2.1. Related Work*

Over the last few years, from 2013 to 2018, human daily activities such as sitting on chairs and on the floor, lying right, lying left, slow walk, brisk walk, walking upstairs, walking downstairs, standing, laying, etc. have been extensively studied. Many researchers have worked towards evaluation of HAR using the smartphone with different daily activities. These activities are classified based on feature extraction schemes that are broadly categorized as time and frequency domains. In [19–25] researchers have implemented the time domain and frequency domain feature extraction as a combined approach. Other researchers in [26–31] have used feature extraction in the time domain only, whilst, in [32] researchers applied the frequency domain and time-frequency domain. The authors in [33] have chosen the time domain, frequency domain and time-frequency domain for feature extraction. Out of these studies, Saha et al. [19] found that the ensemble classifier performs best, with an overall accuracy rate of 94% using accelerometer and gyroscope sensor data. In the research carried out by Mohamed et al. [20], a combination of accelerometer data from the arm, belt and pocket analysed using

rotation forest with the base learner C4.5, was found to provide the best overall classification accuracy rate of 98.9% [20]. Researchers in [21,32], and [23–25] have analyzed the same dataset. Ronao and Cho [21] found that classification using two stages of continuous hidden Markov model (TS-CHMM) achieved the highest overall accuracy rate of 93.18%. Jiang, Yin et al. [32] have obtained the best overall accuracy rate of 97.59% using deep convolution neural network (DCNN). Research reported by Kastner et al. [23] provided the best results with an overall classification accuracy rate of 96.23% with generalized learning vector quantization (GLVQ). Romero et al. [24] have found that One vs. one (OVO), OVO-SVM gives the best overall classification accuracy rate as 96.4%; whilst, Anguita et al. [25] managed to gain a little improvement and reported best overall accuracy rate as 96.5% for classification using One vs. all (OVA), OVA-SVM. Researchers in [26–28], have analyzed the same dataset for walking, jogging, walking downstairs, walking, upstairs, sitting, and standing activities. A study conducted by Sufyan et al. [26] found that classification on voting Multilayer perceptron (MLP) and NBtree give the best accuracy rate for classification based on each activity. This study found that Voting MLP-NBtree gives the best accuracy rate of classification on walking at 99.23%, jogging at 98.86%, walking upstairs at 93.35%, walking downstairs at 90.15%, sitting at 98.37% and standing at 98.37%. Research by Daghistani and Alshammari [27] found the best overall classification accuracy rate using Adaboost (J48) at 94.034%, whilst, Catal et al. [28] reported the highest overall classification accuracy rate based on voting (J48, logistic regression and MLP) as 94.06%. Gupta and Kumar [29] stated the study on human activities on sitting, standing, walking and running produced the best overall accuracy rate of 98.83% using an Adaboost classifier. Research by Gao et al. [33] showed that C4.5 was the best classifier with a 96.4% overall accuracy rate for lying, sitting, standing, walking and transition activity. Bayat et al. [30] studied HAR comparison between different classifiers such as MLP, SVM, RF, simple logistic, logitboost, Logistic model tree (LMT) and voting classifier with the triaxial accelerometer data in a smartphone that placed in pocket and hand. The results of this study showed that data in hand produces best overall classification accuracy rate using voting combination of MLP, logitboost and SVM classifier with 91.15%, whilst, data in pocket gives best accuracy of classification rate as 90.34% using voting combination of MLP, RF and simple logistic. Research studies from Ha and Ryu [31] have reported that the best overall classification accuracy rate of 97.8% was obtained with an ensemble method known as Error correcting output coding (ECOC) that was combined with the random forest as the base learner. Overall, it can be concluded that Ensemble methods produce better results compared with other algorithms.
