Annals of Advanced Biomedical Sciences MEDWIN PUBLISHERS ISSN: 2641-9459 Committed to Create Value for Researchers Smartphone and Smartwatch for Human Activity Recognition Esra’a Alshawwa, Mousa Al Zanina, Mohammed Ibbini and Mashhour M Bani Amer* Research Article Volume 4 Issue 1 Department of Biomedical Engineering, Jordan University of Science and Technology, Jordan Received Date: March 24, 2021 Published Date: May 04, 2021 *Corresponding author: Mashhour M Bani Amer, Department of Biomedical Engineering, DOI: 10.23880/aabsc-16000159 Jordan University of Science and Technology, Irbid, Jordan, Email: [email protected] Abstract Human activity recognition (HAR) systems are developed as aspect of a model to allow continual assessment of human behaviors in IoT environments in the areas of ambient assisted living, sports injury detection, elderly care, rehabilitation, and entertainment and close monitoring. Smartphones are already used to recognize activity. Most of the research done in this field system, to promote the process of classifying raw data from smartphone sensors to human activities. Smartwatches solve this placed a restriction on fixing the smartphone securely in a certain location on the human body, along with the machine learning limitation by placing them in a consistent position, which becomes steady and precisely sensitive to body movements. For this experiment, we evaluate both the accelerometer and the gyroscope sensor on the smartphone and the smartwatch, and decide which sensors hybrid does superiorly. Five daily physical human activities are evaluated using five classifiers from WEKA, in addition to Artificial Neural Network (ANN), K- Nearest Neighbor (KNN), and Support Vector Machine (SVM) algorithms built- The results showed that the accelerometer sensors combination has the highest accuracy among other combinations and in MATLAB 2018a. We used confusion matrix and random simulation to compare the accuracy and efficiency of those models. Keywords:achieved an overall accuracy of 97.7% using SVM that gives the best performance among all other classifiers. Accelerometer; Gyroscope Sensor; Human Activity Recognition; KNN; ANN; SVM Abbreviations: HAR: Human Activity Recognition; health monitoring [2,3], assistance applications, emergency services, and transportation assistance services [4]. It is predicted that smart environments, which communicate WEKA: Waikato Environment for Knowledge Analysis; ANN: Artificial Neural Network; KNN: K- Nearest Neighbor; will be a correlative fraction of everyday life in the short term. SVM: Support Vector Machine; DCT: Discrete Cosine with the disabled or elderly according to their specific needs, Mining;Transformation; MP: Multilayer PCA: Principal Perceptron; Component RF: Random Analysis; Forest; SMA: LR: LogisticSignal Magnitude Regression. Area; WISDM: Wireless Sensor Data TheGiven economic that the consequencesglobal rise in theof international older people senilityratio is canfirmly be mitigatedsignificant, by the encouraging aging in place the has overages been of extreme to remain significance. energetic Introduction and stable in their homes for long years, where living free is more normal and restful [5]. Unfortunately, increasing Human Activity Recognition (HAR) is the challenge of recognizing a physical activity conducted within a given environment by an individual based on a trace of toelderly manage population health by makesoneself it in infeasiblecooperation for with all healthcarehomes of movement. HAR in an automated way is basic in many professionalselderly people is to unavoidable. assign a human caretaker, where the need ambient intelligence applications such as smart homes [1], Smartphone and Smartwatch for Human Activity Recognition Ann Adv Biomed Sci 2 Annals of Advanced Biomedical Sciences The smartphone has recently become an integral part of human regular activities and is turning into an increasingly sophisticated system with rising processing framework is outlined in Section III, the outcomes of the experiments are given in Section IV, and Section V sums up as 25% of people used a smartphone in 2015, and held our conclusionLiterature and Review explores scopes of future research. theircapacity, phone network wherever functionality they went and [6]. sensingIt also includes abilities. a Manylarge range of hardware sensors such as accelerometer, gyroscope, compasses, barometer, temperature, humidity, light sensor, Internet connections and various mobile apps, smartphones and GPS receiver. Mobile health technology is the interchange haveWith become ever an integralgreater partcomputing of our daily capabilities, lives, as described flexible of eHealth and mobile technology [7]. Smartphone sensors above. Furthermore, even cheaper smartphones have a variety of sensors (accelerometer, GPS, and gyroscope, etc.) that allow human activities to be detected using a have turned into a data source to track different human smartphone. Oscar Lara and Miguel [9] Lobrador introduce thesebehaviors, behaviors for example, can be physicalused to activitiesmonitor consumersuch as running, safety walking, walking upstairs, and downstairs. The analysis of the researchWaikato Environmentcommunity of for machine Knowledge learning. Analysis This provides(WEKA), smartphoneand supply health to the care human services body on in time. order The to key get constraint the best enforcementswhich is definitely of many the learningmost well algorithms recognized and resource enables within them performanceimposed by thefrom current the raw systems data of theis tosmartphone firmly attach sensors. the to be readily tested for a dataset using, among others, cross- In reality, this is not a practical solution, because the user often needs to serve for calling, using the internet, social HAR. validation and random break. Therefore, it helps in solving in various places and then being subjected to dramatic movements.media, etc. on This the technique mobile, which involves means the readingskeeping thecontrolled device what we have employed in this project. Zhenyu and Lianwen by inertial sensors on the device, which could then display [10]Most describe articles a high-precision use extractions HAR of method features that different uses discrete from incorrect results. Smartwatches that are worn in a consistent cosine transformation (DCT), Principal Component Analysis location [8], tackle this restriction. used(PCA), and Support Vector Machine (SVM) to distinguish smartwatch is to hold the smartwatch sensors (accelerometer various individual activities. Vollmer C, et al. [11] have and Thegyroscope key technologicalcoordinates) injustification a location that for is usingstable andthe [12] toShift-invariant identify Sparse Coding Algorithm for activity critical to movements of the human body. In other words, running,recognition. and SVM algorithm was implemented by Cho J, et al. measuring walking, moving up stairs, moving down stairs, generalized and distinct features to be utilized for outcomes motionless. The characteristics picked from the thatthe aforementionedare more reliable smartwatch in the system sensors of recognition can fit outof human more system magnitudedata were areastandard (SMA). deviation Additionally, of Y-axis, the activity. Y-axis correlation, Z-axis correlation, Y-axis autoregressive fitting and signal et al. [13] suggested a gait-based Advances in machine and deep learning techniques for mean, standard deviation, and pitch skew were chosen selecting features besides the addition of a variety of sensors waveletfor classification. transformation Boyle, was used to obtain characteristics will move the limits of recognition of human behavior to method for the identification of walking carried behaviors. out using The challenges by collecting and preprocessing available to the miningfrom raw was data, also and used the to identifyclassification and classify was HAR using some deeper epistemological levels. This project tackles the above the K-Nearest Neighbors (KNN) algorithm. Functions,WEKA Lazy,data Meta, Mi, Misc, Rules, and Trees) [14]. Some datasets were recognizepublic benchmark daily life activities.data, after Our which project we use describes machine the learning activity generatedof the classifier by researchers, algorithms submitted,(such as Bayes, and used by others and deep learning to predict and accurately and efficiently Data Mining) is a Fordham University published dataset, which includes recognition task and the procedure for performing this task dataacross gathered the world. under WISDM controlled (Wireless and laboratorySensor conditions using different machine learning algorithms including WEKA [15]. Only the accelerometer installed in the mobile phone is data mining, artificial neural network (ANN), K-nearest used to collect data. The sampling rate is 20Hz and the total andneighbor descending), (KNN), andsitting, support and standing. vector machine (SVM), where we consider five activities: walking, jogging, stairs (ascending The remainder of the article is constructed as follows. instance count is 1,098,207. There are six features (user Section II introduces a literature review of human activity index, type of activity, time stamp, x, y and z-accelerations). recognition. The methodology proposed for the built (11.2%),The collection downstairs task involves (9.1%), twenty-nine sitting (5.5%), subjects. and Sixstanding tasks are classified (walking (38.6%), jogging (31.2%), upstairs Mashhour M Bani Amer, et al. Smartphone and Smartwatch