Intelligent assessment and prediction system for somatic fitness and healthcare using machine learning

Abstract

This study proposes a health assessment and predictive assistance system for intelligent health monitoring. Through machine learning, the tool features a customized set of quantitative measurements and web analysis systems for physical and mental fitness. The system replaces the manpower and time requirements of the past necessary to conduct interviews and keep paper records, allowing users to observe and analyze physical and mental fitness status through the webpage. To achieve this, ECG, EEG, and EMAS are used to follow physiological, psychological, and meridian energy states. ASP.NET software is used as a development tool for the system cloud page, which constructs, documents, evaluates, and predicts functions for the smart health assistance system. The measurement data is entered and recorded in the cloud database. The data is used to construct an assessment and prediction of the user’s state of mind and body through machine learning methods, as well as the individual’s physical and mental fitness.

Keywords

Intelligent assessment intelligent prediction somatic fitness healthcare machine learning

1 Introduction

After observing that electric waves were emitted from the electric eel in 1929, Hans Berger guessed that the same phenomenon would occur in the human body [2]. This was the first time the same radio wave activity was recorded from a human skull and the first time in history that the phenomenon of recorded brainwaves was published. Now known widely as “Electroencephalography” (EEG), Berger not only discovered brainwaves but also described different types of brainwaves. The EEG records the current activity of the brain by placing non-invasive electrodes on the scalp and recording subsequent neurophysiological measurements or by placing invasive electrodes on the brain and measuring cortex signals in special cases. This study served as the basis for the selection of quantitative sensing instruments and increased awareness regarding the brain.

In 1956, when Nakatani Yoshio [10] studied acupuncture points and meridians, it was found that there were 370 “Ryodoten”, which was consistent with the 361 points recorded in traditional acupuncture. After this discovery, medical scientists in numerous countries confirmed the objective foundation of the existence of acupuncture points. Later, Yoshio observed that these kinds of Ryodoten and Ryodoraku are from the internal organs linked by the autonomic nerves (especially the sympathetic nerve), and can be found by measuring skin resistance and current. It can be further observed whether the sympathetic function and the function of the internal organs are normal or abnormal. Through this basic theory, the Electro Meridian Analysis System (EMAS) was developed for acupuncture point detection and diagnosis. This study allowed for the choice of quantitative sensing instruments for measuring health status and changes in internal organs.

In 2005, Ling-Ping Lai [7] proposed Lai’s electrocardiogram. The heart is an important organ for maintaining life, and its health needs to be monitored by instruments. Electrocardiography is currently the least invasive medical method and it is one of the most basic tests in the diagnosis and treatment of cardiovascular diseases. The principle is that the weak current generated by the contraction and expansion of the heart is sensed by the electrodes attached to the epidermis, transduced into a signal by the machine, and recorded on a paper strip by corrugation, resulting in what is known as an electrocardiogram (ECG). This work enabled the selection of quantitative sensing instruments and formed the basis for measuring cardiac function.

In 2008, Toh, Foh-Fook [12] used the meridian energy diagnostic instrument to measure the average meridian energy of sub-optimal health conditions in the liver, gallbladder, heart, small intestine, spleen, stomach, lung, kidney, bladder, or in cases of anxiety. Three meridians are of normal values if they are between 40 and 70. Meridian values in this range indicate that body energy and visceral functions maintain a relative balance and coordination. A value greater than 70 indicates that the body is in a state of overabundance, and a value of less than 40 indicates that the body is in a weak state. The average energy was used to understand the overall meridian state of the patient and the evaluation of data provided by the meridian detector was used for follow up analysis in this study.

In 2008, Yu-Han Hung and Ching Li [6] proposed five stages of construction elements to organize a fitness management system, which was made up of the detection project, evaluation methods, diagnostic criteria, exercise prescriptions, and activity interventions for physical fitness management. In addition, through personal and professional health and medical assistance, an exercise strategy could be planned. This study developed the five stages of the institute’s design for physical and mental fitness.

In 2009, Terrance Malkinson [9] explored the roles of emerging technologies in endurance sports training and competition. Techniques for monitoring athlete health during a triathlon (swim, cycling, and running) included the use of transponders, GPS, power measurement, kinematics, biology mechanics, wearable sensors, virtual aids, and testing. The data obtained from these techniques during exercise training, as well as that obtained during sports events, provides valuable information to the exercise participant, predicting injuries and monitoring performance.

In 2009, Ahtinen et al. [1] reported the user experiences of mobile wellness applications Health Diary, Mobile Coach, and Self Relax for health promotion. The ongoing Nuadu trial investigates user acceptance and technical effectiveness in supporting working-age citizens’ health management. These technologies consist of different mobile web apps and wearable solutions. User experiences from three mobile health applications (Health Diary, Mobile Coach, and Self Relax) were introduced. The usefulness, perceived availability, usage habits, and incentives were collected during the first phase of the trial. User experience data, conducted surveys, and interviews, as well as actual usage logs for mobile applications were collected. The results showed that habits become more practical during the two-month period of use. The results point to several aspects to consider in future health applications including adaptability, versatility, guidance, and usability. This article outlines the needs and required features of building mobile coaching web applications.

In 2010, Wen-Chuan Xie, Jun-Ting Ding, et al. [13] proposed a self-made biomedical sensor, improved over the ANT wireless sensing network standard, that integrated the concept of health promotion by combining bicycle riding and stepping motion sensing information. The sensor sent data to the three mainstream operating systems: Windows Mobile, iPhone OS, and Android portable platform as an analytical gateway to provide timely functioning. In addition, through the 3G network, the information analyzed by the mobile platform is synchronized to the Google app engine cloud, providing chronic disease warnings to health banks. Integrating cloud computing technology and wireless sensing capabilities into medical care and health promotion services can provide a new solution for the management of health system platforms.

In 2012, Bosems [3] proposed a ubiquitous system for the intelligent monitoring of family and work well-being to combat the decline of lifestyle in the Netherlands. Unhealthy habits cause physical and psychological problems. Therefore, in order to understand both physical and psychological problems, wireless sensors were proposed to obtain data from users and transfer it to mobile devices (mobile phones, PDAs) or fixed devices (laptops, computers). The system visualizes the data and uses inference algorithms to provide user recommendations. The device will transfer and share the acquired data so that it can be linked to other available data to compile a database, which can be further used to enhance recommendations. In this paper, the application of sensor transmission to a database is proposed, but the subsequent application of analysis was not used. Therefore, the application of machine-based intelligent health assistance systems has been proposed.

In 2013, Jun-Yuan Chen et al. [4] from Taiwan’s physical and mental motion educational program presented an analysis based on sports and sports research. The researchers are educators, so they integrated experimental methods and interviews simultaneously to gather data. Through the analysis of written and filmed interview records and teaching notes, the measurer devises a scale for the experimental group to compare with the control group. The scale is derived from a questionnaire featuring different research questions that allow for the collection of data and statistical analysis. However, most of the subjects cannot track the data during the measurement, and there may be problems such as data loss or errors through this method. Still, this research could effectively achieve long-term tracking results through cloud computing records.

In 2017, Qingxue Zhang et al. [14] proposed a fully ear-worn long-term blood pressure (BP) and heart rate (HR) monitor to achieve a higher wearability using a machine learning framework to deal with severe motion artifacts induced by head movements. Electrocardiogram(ECG) and photoplethysmography(PPG) sensors are situated behind two ears to achieve a good wearability, and weak ear-ECG/PPG signals are successfully acquired using a semi-customized platform. dealing with many severe motion artifacts, the raw heartbeats are identified using the SVM classifier, and then distorted or faking raw heartbeats are automatically purified by an unsupervised learning algorithm. It is expected to demonstrate the feasibility of the proof-of-concept system in wearable ear ECG/PPG acquisition and motion-tolerant BP/HR estimation, to enable pervasive hypertension, heart health and fitness management. In future, data will be acquired from more subjects, and also further motion artifacts will be introduced from more scenarios, such as walking, running, sleeping, eating, etc.

In 2018, Lichen Ma [8] proposed a kind of multi-level estimation and fuzzy evaluation of physical fitness and health effect of college students in regular institutions of higher learning based on classification and regression tree algorithm (CRTA). In view of the problem that it is impossible to effectively evaluate the multi-level estimation of college students in regular institutions of higher learning with the data theory, the CRTA is applied to carry out pretreatment on the physical fitness and health of the college students in the regular institutions of high learning. In accordance with the characteristics of the CRTA, calculation is performed on the physical fitness and health test data of the college students. The experimental results show that CRTA method can make multi-level estimation and fuzzy evaluation of the physical fitness and health of the college students in the regular institutions of high learning both reasonably and efficiently. For the problem of the uncertainties in the multi-level estimation fuzzy evaluation of the college students in the regular institutions of higher learning, the massive data that have been collected by using the CRTA, so as to form a number of clusters. Then the physical fitness and health test data of the college students in all the clusters are calculated as the representative of the clusters. From the experimental results of the simulation experiment, it showed that: The CRTA used in this paper can ensure that the time for the evaluation of the massive information can be effectively reduced under the premise that correct evaluation results are obtained. It saved the time and achieved high efficiency. So it is a highly efficient and reasonable method for the multi-level estimation and fuzzy evaluation of physical fitness and health effect of college students in regular institutions of higher learning in the new era.

In 2019, A. Depari et al. [5] proposed a new approach for automatic classification and counting of workout exercises. Machine Learning techniques of PCA and LDA have been employed in order to reducing the computational efforts, thus assuring a possible implementation in low-cost and resource-constrained wearable devices. The data set was obtained monitoring the workout of seven users with different capabilities. It offers a great variability in performance by the exercise classification, but it assures an accuracy over 93% and the exercise counting feature shows a maximum average error of 6%. With positive results, an evaluation of possible implementation in embedded devices has been also provided.

Currently, research is being carried out to integrate medical technologies such as cloud computing and wireless sensing networks into medical care and health promotion services. There has been related research for assessing and predicting the state of mind and body, but further studies are necessary. Although there have been psychological explorations examining brain waves during physical activity, these systems are rarely extended to the application of “Somatic Fitness”.

In 2009, William B. Strean et.al [11] illustrated how a somatic view creates potentially powerful methods of working through the body, enabling a more holistic approach to understanding and helping clients. They proposed a case study to show how these perspectives are incorporated into initial consultations, intakes, and interventions. Therefore, this study constructs an all-in-one intelligent somatic fitness system for health assessment and prediction.

2 Design and implementation of research methods

The purpose of this study was to develop and construct a “health assessment and prediction system for the intelligent monitoring of fitness”, which is divided into two parts: (1) The intelligent health assessment system and (2) The intelligent health prediction system. As shown in Fig. 1, the wireless detection instrument that is suitable for the dust-end can capture the physical fitness related data, transmit it to the cloud database through the microcontroller’s server and then build the user-interface using the ASP.NET framework. The interface is also used by the intelligent inference engine to learn the user’s physical fitness data through various algorithms as a systematic data analysis operation. In the process, the evaluation system conducts an initial health assessment from data obtained in the first step and infers trends through the prediction system.

Fig. 1

Overview of the intelligent health assistance system.

The system architecture is shown in Fig. 1. Body and mind fitness data are collected and sent to the cloud database for access through the dust end sensor instruments (ECG, EEG, and EMAS). The machine learning captures the cloud data as a sample and builds a model to generate a WEB application package to the web page for back-end development. The user can use the 3C product to access the webpage through the network to observe and analyze their physical fitness status. Through the intelligent health-assisted recording system, the functions of observing, adding, modifying, and deleting records can be carried out in the process. This intelligent fitness-assisted evaluation system learns from the fitness scores and infers trends of personal fitness.

2.1 Method design

2.1.1 Intelligent health assessment system

This study used six input feature parameters for SFEI (Somatic Fitness Evaluation Index) customization, as shown in Table 1, we define the assessment of physical fitness as: If there are more than four ‘poor’ indicators, the assessment of physical and mental state is poor; if there are more than four ‘good’ indicators, the physical and mental status is good; if more than three indicators are moderate or good, the physical and mental state is evaluated as moderate. This study uses Azure Machine Learning Studio to perform multi-class classification algorithm screening and comparison. The process of comparing multiple steps using neural networks, multi-type decision-making jungles, multi-class logistic regression, and multi-class decision forest processing is shown in Fig. 2. The sample size of the experimental group with 7 males and 1 female of 20±1 years old is 100. First, data is entered into the Azure Machine Learning Studio design environment by selecting 100 SFEI data samples for HRV, Stress, Mood, Attention, Meditation, AVG, SFEI and other indicators to enter the data segmentation module to segment the data into 75% training and 25% evaluation samples. The four algorithms are used to select the attributes and the training module is selected to train the SFEI parameters. The score module can observe the evaluation samples and learn the training results through the training samples. Finally, the evaluation module can present the comprehensive precision and average accuracy for the screening process.

Table 1
Design parameters of the intelligent health assistance system for the assessment of fitness level

Score Definition\Parameter Index HRV Stress Mood Attention Meditation AVG SFEI

Poor 0–40 71–100 71–100 0–40 0–40 0–20 1

86–100

Moderate 41–70 41–70 41–70 41–70 41–70 21–40 2

71–85

Good 71–100 0–40 0–40 71–100 71–100 41–70 3

Score Definition\Parameter Index	HRV	Stress	Mood	Attention	Meditation	AVG	SFEI
Poor	0–40	71–100	71–100	0–40	0–40	0–20	1
						86–100
Moderate	41–70	41–70	41–70	41–70	41–70	21–40	2
						71–85
Good	71–100	0–40	0–40	71–100	71–100	41–70	3

Fig. 2

The flow chart of multi-class classification algorithm screening and comparison for intelligent health assessment system.

2.1.2 Intelligent health prediction system

The intelligent health prediction system, as shown in Table 2, takes the six input parameters of the table (HRV, Stress, Mood, Attention, Meditation, and AVG) and outputs the parameter SFPI (Somatic Fitness Prediction Index). Using regression algorithms, we can speculate on the future fitness of the user and the unknown state of mind and body through the parameter rules that have been determined in this study. After the user measures a new data, the model can determine whether or not the data is moving toward a better physical and/or mental state. The intelligent health assistance system predictive model uses Azure Machine Learning Studio to compare regression type algorithms. Specifically, the Bayesian Linear Regression, Boosted Decision Tree Regression, Decision Forest Regression, Linear Regression, Neural Network Regression, and Poisson Regression are used in the process. Figure 3 illustrates the comparison step process via a flow chart of the regression prediction calculation for the model. The process begins by entering the Azure Machine Learning Studio design environment, selecting 100 SFPI data samples for HRV, Stress, Mood, Attention, Meditation, AVG, and SFPI and other indicators into the data segmentation module to be divided into training samples (70%) and evaluation samples (30%). Then the regression algorithm is selected from Bayesian Linear Regression, Enhanced Decision Tree Regression, Decision Forest Regression, Linear Regression, Neural Network Regression, and Poisson Regression Law. The training module is selected to train the SFPI parameters, and the score module can observe the results of the evaluated samples and the training results after learning through the training samples. Finally, the evaluation module can present the results of the average absolute error and screen the root mean square error. The sample size of the experimental group with 7 males and 1 female of 20±1 years old is 100.

Table 2
Design parameters of the intelligent health assistance system for the prediction of fitness level

Index Score Definition\Parameter HRV Stress Mood Attention Meditation AVG SFPI

Regress 0–40 71–100 71–100 0–40 0–40 0–20 1±0.5

86–100

Maintain 41–70 41–70 41–70 41–70 41–70 21–40 2±0.5

71–85

Progress 71–100 0–40 0–40 71–100 71–100 41–70 3±0.5

Index Score Definition\Parameter	HRV	Stress	Mood	Attention	Meditation	AVG	SFPI
Regress	0–40	71–100	71–100	0–40	0–40	0–20	1±0.5
						86–100
Maintain	41–70	41–70	41–70	41–70	41–70	21–40	2±0.5
						71–85
Progress	71–100	0–40	0–40	71–100	71–100	41–70	3±0.5

Fig. 3

The flow chart of multi-class classification algorithm screening and comparison for intelligent health prediction system t.

2.2 Method implementation

2.2.1 Intelligent health assessment system

After the user logs into the page, they enter the smart health assistance diagnosis system and can input HRV, Stress, Mood, Attention, Meditation, AVG, and SFEI parameters for the multi-type decision tree forest algorithm. At this time, after the algorithm captures the sample SFEI data and performs the training and model construction, the user can observe the output result of the SFEI. The user can select whether to record the data, so that they can access the calculated results in the SFEI sample data table in the training database. In doing so, the next time the training is performed, the SFEI can more accurately infer the output of the physical and mental assessment. The steps of the process are shown in Fig. 4 and 5.

Fig. 4

Diagram for the assessment of the intelligent health assistance system’s model architecture.

Fig. 5

The implementation flow chart of intelligent health assessment system.

The forest training learning method for the intelligent health assistance model is described as follows:

Collect 100 samples of physical and mental assessment data. In the process, 70% will be used as the training set and 30% will be used as the test set. The resampling method will be used for bagging. The decision tree number is 8 and the maximum depth of the decision tree is 32. The minimum number of leaf node samples is 1.

Inside the intelligent health assistance evaluation system, the measured data is sequentially input as shown in Fig. 6 as an example with HRV input 20, Stress input 30, Mood input 40, Attention input 50, Meditation input 60, AVG input 40. After the values are filled in, click the operation button to get the assessment of physical and mental status for your health being moderate. Upon pressing the record button, the value of the SFEI is transmitted back through the machine learning operation, and the input value is stored in the SFEI database. After the record button is selected, the value of the input box will be cleared and a box will pop up saying that it has been successfully recorded, as shown in Fig. 7.

Fig. 6

Intelligent health assistance system assessment page.

Fig. 7

Intelligent health assistance system assessment record page.

2.2.2 Intelligent health prediction system

The intelligent health assistance prediction system is used to enter the HRV, Stress, Mood, Attention, Meditation, AVG, and SFPI parameters through the user login page to be used in the decision forest algorithm. At this time, the algorithm can capture the sample SFPI data and construct the training model, then the user can observe the output results of the SFPI. It can then be chosen whether to record the data, so that users can access the calculated data in the SFPI data table in the training database for the next training. The SFPI output can be more accurately inferred to predict the user’s state of mind and body. The process is shown in Fig. 8 and 9.

Fig. 8

Diagram for the prediction of the intelligent health assistance system’s model architecture.

Fig. 9

The implementation flow chart of intelligent health prediction system.

The decision tree calculation using regression learning for the intelligent health assistance evaluation model is described as follows:

For verifying the prediction model, new 206 sets of input and output samples of body and mind prediction data are collected. In the process, 70% of the training set will be used as the test set, and the maximum number of each leaf node is 20, and the minimum number of samples per leaf node is 10. The learning rate is 0.2. The total number of constructed trees is 100 nested into the decision tree calculus.

Inside the intelligent health assistance prediction system, the measured data is input in sequence as shown in Fig. 10 as an example with HRV input 20, Stress input 30, Mood input 40, Attention input 50, Meditation input 60, AVG input 40. After the values are filled in, click the operation button to get the predicted state of mind and body. Upon clicking the record button, the value is transferred back to the SFPI after the machine learning operation, and the input value is stored in the SFPI database. As shown in Fig. 11, the value of the input box will be cleared and the successfully recorded notification will be given.

Fig. 10

Intelligent health assistance system prediction operation page.

Fig. 11

Intelligent health assistance system prediction record page.

3 Results and discussions

3.1 Intelligent health assessment system

The intelligent health assistance evaluation model gives a quantitative assessment of the physical and mental fitness of the body and mind through the collection of data using the meridian detector, ECG, and EEG. This allows the user to understand his or her physical and mental condition through quantitative data. The assessment results are divided into three categories: poor, moderate, and good. The main purpose of using classification algorithms for screening is to let the subject know clearly which algorithm is the most suitable as the design model of the evaluation system. Therefore, the model algorithm uses multiple types of decision forests to derive the SFEI body and mind fitness assessment. The average accuracy of using the multi-class decision tree forest method is 0.88 with a max accuracy of 0.92. The accuracy rate is checked through the output results. The correct rate of the model evaluation results are as follows: predicted indicators/predicted subjects overall receive 88% of the correct rate results output as shown in Table 3.

Table 3
Correct rate of intelligent health assistance assessment system

Specification Project\Subsystem Evaluation

Subsystem

Forecast indicator 22

Predicted subject overall 25

Correct rate 88%

Specification Project\Subsystem	Evaluation
Forecast indicator	22
Predicted subject overall	25
Correct rate	88%

3.2 Intelligent health prediction system

The intelligent health assistance prediction system model is used to predict the physical and mental fitness status through the results provided by the pre-evaluation step of the intelligent health assistance system, and to distinguish the three physical and mental predictive states of the regression, maintenance, and progress using the predicted results. The regression algorithm is used for screening. The main purpose is to let the subject know more clearly about the trends of their physical and mental fitness. In the process, the mental health variability, stress index, concentration, meditation, and overall meridian physical and mental variables (HRV, Stress, Mood, Attention, Meditation, and AVG) were used to evaluate SFPI fitness for the body and mind. The algorithm uses the Decision Forest Regression method and returns an average absolute error of 0.130139. The root mean square error is 0.243471, as shown in Table 4. However, there are errors in the numerical display. Therefore, the back-end SFPI tolerance range value is compared with the model prediction result by the prediction subsystem. The evaluation criterion is the prediction correct rate (prediction index/predicted project overall). By comparing the model prediction results with the actual tolerance value, with a±0.5 adjustment, the correct rate of the prediction model can achieve 80% of the correct rate result output. The correct rate of the prediction model obtained after entering the back end for tolerance adjustment can reach 95% as shown in Table 5. And the way to calculate correct rates in Tables 3 and 5 are using Azure Machine Learning Studio. So the Decision Forest Regression method is selected for the prediction system.

Table 4
Comparison table of regression prediction algorithms for intelligent health assistance system prediction model

Indicator\Algorithm Bayesian Linear Regression Boosted Decision Tree Regression Decision Forest Regression Linear Regression Neural Network Regression Poisson Regression

Mean Absolute Error 0.294205 0.151919 0.130139 0.297866 0.327859 0.311261

Root Mean Square Error 0.37182 0.330218 0.243471 0.396097 0.427199 0.435319

Indicator\Algorithm	Bayesian Linear Regression	Boosted Decision Tree Regression	Decision Forest Regression	Linear Regression	Neural Network Regression	Poisson Regression
Mean Absolute Error	0.294205	0.151919	0.130139	0.297866	0.327859	0.311261
Root Mean Square Error	0.37182	0.330218	0.243471	0.396097	0.427199	0.435319

Table 5

Correct rate of intelligent health assistance prediction system

Specification Subject\Subsystem	Predicted Model	Predicted Subsystem
Predicted Indicator	48	57
Predicted Project Overall	60	60
Correct Rate	80%	95%

4 Conclusions

This study developed a new tool based on machine learning as a smart health assistance system. Experiments showed suitability as an application for the quantification of the state of physical and mental fitness. Through the actual measurement of physical and mental fitness parameters, the smart health assistance system proposed in this study can indeed be verified with a model accuracy rate of over 88%. The maximally accurate (95%) intelligent health aid system prediction model was developed using a solid state system architecture.

Through the intelligent health assistance system, the tool can evaluate and predict the physical and mental fitness state of users, who can follow as the machine learning approach develops and constructs a model health plan for the body and mind. The physical and mental fitness can then be used to derive future trends by capturing personal data on physical and mental fitness. If the test only had a simple date record, it would be necessary to conduct statistical analysis by professionals to determine the trends of physical fitness. Therefore, our system can assist the needs of general users.

Footnotes

Acknowledgments

We would like to thank the Ministry of Science and Technology of the Republic of China (Taiwan) for financial support of this research under contract numbers MOST 104-2622-H-269-002-CC3.

References

Mattila

A.E.

, Vaatanen

, Hynninen

, Salminen

, Koskinen

and Laine

, User experiences of mobile wellness applications in health promotion: User study of Wellness Diary, Mobile Coach and SelfRelax, in: 2009 3rd International Conference on Pervasive Computing Technologies for Healthcare, IEEE, 2009, pp. 1–8.

Berger

, Über das elektrenkephalogramm des menschen, European Archives of Psychiatry and Clinical Neuroscience 87 (1929), 527–570.

Bosems

, A ubiquitous system for smart reasoning for well-being at home and atwork, in: 2012 IEEE International Conference on Pervasive Computing and Communications Workshops, IEEE, 2012, pp. 534–535.

Chen

, Liu

and Chang

, Analysis of Somatic Movement Education Applied in Physical Education and Sports in Taiwan (2002 2012), Body Culture Journal (2013), 1–17.

Depari

, Ferrari

, Flammini

, Rinaldi

, Sisinni

, Lightweight Machine Learning-Based Approach for Supervision of FitnessWorkout, in: 2019 IEEE Sensors Applications Symposium (SAS), IEEE, 2019, pp. 1–6.

Hung

and Li

, The Study of the Elements to Establish a Physical Fitness Management System, MingDao Journal of General Education (2008), 89–102.

Lai

, Lai’s Electrocardiogram, Yi Hsien Publishing Co., Ltd, 2005.

, Multi-level Estimation and Fuzzy Evaluation of Physical Fitness and Health Effect of College Students in Regular Institutions of Higher Learning Based on Classification and Regression Tree Algorithm, in: 2018 International Conference on Virtual Reality and Intelligent Systems (ICVRIS), IEEE, 2018, pp. 201–205.

Malkinson

, Current and emerging technologies in endurance athletic training and race monitoring, in: 2009 IEEE Toronto International Conference Science and Technology for Humanity (TIC-STH), IEEE, 2009, pp. 581–586.

10.

Nakatani

, Skin electric resistance and ryodoraku, J Autonomic Nerve 6 (1956).

11.

Strean

and Strozzi-Heckler

, Body Knowledge: Somatic Contributions to the Practice of Sport Psychology, Journal of Applied Sport Psychology 21 (2009), 91–98.

12.

Toh

, The Evaluation of Sub-health through the Meridian Energy Diagnostic Equipment, Science Journal of Taiwan Traditional Chinese Medicine 3 (2008), 11–19.

13.

Xie

, Ding

, Chien

and Weng

, The Research and Manufacturing of Wearable Health Management System Based on Cloud Operation, National Kaohsiung University of Applied Sciences (2010).

14.

Zhang

, Zeng

, Hu

and Zhou

, A machine learning-empowered system for long-term motion-tolerant wearable monitoring of blood pressure and heart rate with ear-ECG/PPG, IEEE Access 5 (2017), 10547–10561.

Specification Project\Subsystem	Evaluation
	Subsystem
Forecast indicator	22
Predicted subject overall	25
Correct rate	88%