Journal of Physics: Conference Series

PAPER • OPEN ACCESS You may also like

- Estimation of size of red blood cell
An IoT Based diabetic patient Monitoring System aggregates using backscattering property
of high-frequency ultrasound: In vivo
Using Machine Learning and Node MCU evaluation
Yusaku Kurokawa, Hirofumi Taki, Satoshi
Yashiro et al.
To cite this article: Amine Rghioui et al 2021 J. Phys.: Conf. Ser. 1743 012035 - Factors influencing breath analysis results
in patients with diabetes mellitus
Qing Zhou, Qiannan Wang, Bo Chen et al.

- Noninvasive monitoring of blood glucose

concentration in diabetic patients with
View the article online for updates and enhancements. optical coherence tomography
Y T Lan, Y P Kuang, L P Zhou et al.

This content was downloaded from IP address 190.141.98.232 on 16/05/2022 at 16:44

The International Conference on Mathematics & Data Science (ICMDS) 2020 IOP Publishing
Journal of Physics: Conference Series 1743 (2021) 012035 doi:10.1088/1742-6596/1743/1/012035

An IoT Based diabetic patient Monitoring System Using

Machine Learning and Node MCU

Amine Rghioui , Assia Naja , Jaime Lloret Mauri , Abedlmajid Oumnad

1 2 3 1

Research Team in Smart Communications-ERSC Research Centre E3S, EMI,

1

Mohamed V University in Rabat, Morocco

2
International University of Rabat, FIL, TICLab, Morocco
3
Integrated Management Coastal Research Institute, Universitat Politecnica de
Valencia, 46370 Valencia, Spain.

[email protected], [email protected], [email protected],

[email protected]
Abstract. Diabetic patient monitoring is a systematic method that provides us with detailed
information about the diabetic patient. Diabetic patient monitoring systems play a significant
role in monitoring the patient's health, especially with the use of Internet of Things (IoT) devices.
Diabetic patient monitoring systems are able basically to monitor diabetic patients and save some
data about blood glucose level, body temperature, and location. The role of this system is not
limited to patients monitoring, it can also classify data using machine learning techniques.
Predictive analytic for diabetic patients is very important due to its ability to help diabetic
patients, their families, doctors, and medical researchers to make decisions on diabetic patient
treatment based on big data. This paper describes a new system for monitoring diabetic patients
and discusses predictive analytics using four different machine-learning algorithms. The
performance and accuracy of the applied algorithms are discussed and compared to choose the
best one in terms of several parameters.

1. Introduction
The technologies of the Internet of objects and their applications have made a great development by
becoming in these days more accessible and more available, allowing a large number of objects to be
interconnected via the Internet in several fields that are the field of health, home automation, industrial
manufacturing, etc [1]. In the field of intelligent health, there are several applications that aim to improve
care and improve the quality of life of patients with chronic diseases. Using IoT, the mobile health
service becomes more important as it plays a very important role in monitoring and controlling patients
who suffer from chronic diseases such as cardiovascular disease and diabetes [2][3]. In the field of
intelligent health and more precisely in the part of patient monitoring, we find that patient data is very
useful. Indeed to realize an IoT application in this field, the one must have assured the recording of a

Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
The International Conference on Mathematics & Data Science (ICMDS) 2020 IOP Publishing
Journal of Physics: Conference Series 1743 (2021) 012035 doi:10.1088/1742-6596/1743/1/012035

large amount of data collected by using measurements of the medical signs on the patients. Analyzes
can be applied to identify people who need "proactive care" to avoid worsening their condition. For
example, patients in the early stages of certain diseases (for example, heart failure often caused by
certain risk factors such as hypertension or diabetes) should be able to benefit from preventive care
thanks to big data [4]. To improve the health of patients, some of them shared their private details to
save lives in return [5].
Nowadays, there are several chronic illnesses, for instance, heart disease, stroke, cancer, chronic
respiratory disease, and diabetes. This is a dangerous disease that is lately becoming one of the leading
causes of death in the world, and which requires a lot of careful monitoring to keep patients healthy.
Diabetes are caused by insulin resistance, and insufficient insulin production can lead to either an
increase or decrease in the level of glucose in the blood, therefore the main challenge of the diabetic
patient is to maintain the glucose level stable within a specific interval. If they can no longer comply
with these conditions, some patients require urgent care to avoid worsening [6].
The growth of diabetic patients in the world implied an increase the use of continuous glucose
monitoring devices (CGM) to control diabetic patient, these devices become the new method of
continuous monitoring. They provide real-time information about the glucose level.
In this article, we present an intelligent system for monitoring diabetic patients using Node MCU
and Machine Learning algorithms. The MCU node is connected with the glucometer to periodically
record the glucose level in a diabetic patient. Using this collected data, patients can be monitored
remotely by caregivers (patients, researchers, and doctors). As a result, patients and doctors alike have
to process multiple records and interpret the huge amount of data to adjust insulin doses and keep blood
sugar as close to normal as possible. An intelligent algorithm is implemented in our system, which is
able to send the data to caregivers, to store them in a database as positive or false positive after being
validated (or not) by the doctor. In our case study, the included sensors facilitate the monitoring of
diabetic diseases.
The rest of this paper is organized in the following: Section 2 presents the related work. Section 3
describes our proposed diabetic monitoring system. Section 4 details analyses Data Pre-Processing.
Section 5 describes the experiments’ setup, as well as the results of our practical evaluations. Finally,
Section 6 concludes this paper.

2. Literature Review
This section discusses the literature review analysis of some papers in relation to mobile application for
diabetic patient monitoring using machine-learning algorithms. A description of the proposed systems
and the used the used algorithms in this work are given. In their work, the authors presented an intelligent
architecture for the surveillance of diabetic disease that monitor the health of diabetic patients through

2
The International Conference on Mathematics & Data Science (ICMDS) 2020 IOP Publishing
Journal of Physics: Conference Series 1743 (2021) 012035 doi:10.1088/1742-6596/1743/1/012035

sensors integrated into smartphones [7]. In another work, the prediction of the diabetes types using
analysis algorithms and Hadoop map-reduce, prediction of complications, and the prediction of the type
of treatment have been investigated [8], while in [9], the authors proposed a system that can perform
predictions for Mellitus which is a type of diabetes using Hadoop/MapReduce. A new system for the
prediction of the glucose concentration has been proposed in [10]. Where the data generated by the
Continuous Glucose Monitoring can be analyzed by the glucoSim software using the Kalman Filter (KF)
to reduce the noise.
Regarding Many researchers have developed and implemented various analysis and prediction
models using different data mining techniques. In [11], the authors used a classification technique with
Naive Bayes and Decision Tree algorithms by using the Weka tool to find out patterns from the diabetes
data sets.
In [12] authors use Naïve Bayes and Decision Trees in one model classification technique to study
hidden patterns in the diabetes dataset. Authors in [13] used the C4.5 decision tree algorithm, Neural
Network algorithm, K-means clustering algorithm, and Visualization to predict diabetes for patients.
In this paper, we are using four machine learning algorithms, Naïve Bayes, Random Forest, OneR,
and J48 classification algorithm from the diabetes data set, to test the most powerful algorithm to
determine the patient’s level of risk. No one of the revised systems is specially designed and developed
for diabetic patient monitoring. This work develops a new application using IoT for monitoring diabetic
patients via Wi-Fi, but we are not limited to the monitoring, but we added the application of machine
learning algorithms to classify the data generated by sensor and recovered by our application to classify
and analyze it.

3. Diabetic Patient Monitoring System

In this part, we try to show our proposed system. It is a diabetic patient monitoring system based on
node MCU. This system consists of three parts, the sensor part for data collection, the data analysis part,
and the processing of the collected data. The diabetic patient monitoring system measures glucose level
for the diabetic patients using glucose sensors and the data captured transmitted using Node MCU with
the help of the IoT platform to the database for storage and treatment. The measured data are classified
and analyzed by machine learning algorithms. The result of the classification is sent to the doctor to
check the measured value, while a message is sent to the patients in an emergency case.

3.1. General description:

The general description of our proposed system for monitoring diabetic patients is presented in Fig. 1.

3
The International Conference on Mathematics & Data Science (ICMDS) 2020 IOP Publishing
Journal of Physics: Conference Series 1743 (2021) 012035 doi:10.1088/1742-6596/1743/1/012035

Figure 1. System Overview.

3.1.1. Glucose Sensor

A glucose sensor is a sensor intended for measuring the glucose level in a patient's blood. It has the same
function as a continuous glucose monitoring system (CGM) and which can be either a sensor external
placed on the skin or inserted under the skin.

3.1.2. Node MCU

Node MCU is a hardware and software IoT platform that integrates a 32-bit microcontroller. Composed
of hardware in the form of an ESP8266 system on an ESP8266 chip manufactured by Expressive
System. It is equipped with a Wi-Fi module, and which interfaces with the system using the USB cable
for loading the program into the laptop.

3.1.3. Cloud Storage

The data collected by the glucose sensor is sent by the Node MCU and stored in the cloud and displayed
on the mobile app. The data is also analyzed by machine learning algorithms and processed in the server.
It is important to keep patient information as it may be used in the future. This database helps the doctor
to interpret and interact in an emergency, in order to give the best and fastest diagnosis.
In this study, we develop a monitoring system for diabetic patients by using Node MCU to help
diabetic patients manage their chronic illnesses themselves. The proposed system records glucose levels
in diabetic patients and sends them automatically to the mobile app to save them on the cloud. The server
can parse and process the data received to classify it. It can be considered as a platform for information
on the level of glucose that allows interaction between patients and doctors. Using the learning machine
algorithms installed in the server, we can automatically help users monitor their glucose levels and
predict future changes in health. Fig. 2 shows the proposed system for diabetic patient monitoring.

4
The International Conference on Mathematics & Data Science (ICMDS) 2020 IOP Publishing
Journal of Physics: Conference Series 1743 (2021) 012035 doi:10.1088/1742-6596/1743/1/012035

Figure 2. Proposed System

4. Data Pre-processing
Due to the serious impact of diabetes on human health, it is relevant to predict diabetes early to better
manage the disease. Artificial intelligence and machine learning algorithms are used widely for data
classification and mainly for prediction. In our paper, we use machine learning to predict the state of
diabetes in diabetic patient. Using machine learning, one can only predict with the data collected from
the glucometer whether the diabetic patient is in the normal state or not without making the direct
medical diagnosis. Therefore, our proposed system uses a supervised classification machine learning
approach that captures the data set collected by the glucose sensor and makes the final decision on
diabetes. Fig. 3 shows the proposed model.

Figure 3. Proposed Methodology Flowchart

5
The International Conference on Mathematics & Data Science (ICMDS) 2020 IOP Publishing
Journal of Physics: Conference Series 1743 (2021) 012035 doi:10.1088/1742-6596/1743/1/012035

4.1. Dataset and Attributes

In this part, we try to give the description of the dataset used as input for the classifiers implemented
using various algorithms. Classification of data is an important step in this work that aims to classify the
data as normal or affected by diabetes. The data studied included the records of 62 diabetic patients (44
men and 18 women) for 67 days with an average of three measurements per day. In this study, we used
4 classification algorithms (Naive Bayes [14], Random Forest [15], OneR [16], and SMO (Sequential
Minimal Optimization)) [17]. In this study, the glucose level data were classified using multiple
classification algorithms in the same data-mining programs. The glucose level data set contains 5
attributes and 12612 glucose level records. Fig. 4 shows blood glucose monitoring in diabetic patients
three times a day, and Table 1 shows the type of data and description of the attributes.

Figure 4. Screenshot of a patient’s stored glucose level

Table.1. Glucose Data types

N° Attributes Data type
1 Sex Boolean
2 Age Digital
3 Day Digital
4 Glucose level Digital
5 BGL Boolean

4.2. Weka Tool Description

To test the four classification algorithms that we used, we worked with the WEKA software (Waikato
Environment for Knowledge Analysis: it is an open-source data extraction program), it is a toolbox of
very popular machine learning and data mining to conduct data-driven research, is a data mining tool
that allows us to classify data, analyze and determine the prediction accuracy of different mining
algorithms. In this study, glucose level data was classified using multiple classification algorithms in
the same data mining programs.

5. Experimental Results

6
The International Conference on Mathematics & Data Science (ICMDS) 2020 IOP Publishing
Journal of Physics: Conference Series 1743 (2021) 012035 doi:10.1088/1742-6596/1743/1/012035

In this section, we treat the experimental results obtained after the simulation of diabetic data with the
classification algorithms Naive Bayes, Support Vector Machine, Random Forest, and Simple CART.
The purpose of this simulation is to evaluate the performance of the classification algorithms that we
use and to recommend the best suitable algorithm for prediction. Using the confusion matrix, we
evaluate the prediction results using various evaluation measures such as the accuracy, classification
accuracy, sensitivity, and the F-Measure.
Confusion matrix: is a matrix composed of the true value of a positive rate (TP), the true value of a
negative rate (TN), the false value of a positive rate (FP), and the false value of a negative rate (FN).
Table2. Confusion Matrix
Positive (1) Negative (0)
POSITIVE (1) TP FP
NEGATIVE (0) FN TN
Precision: It is the ratio of predicted positive instances and the total of all predicted positive instances.
Table. 3 represents the experimental classification accuracy results and the training time of different
algorithms.
Table.3. Correctly and Incorrectly Classified Instances
Algorithms Correctly Classified Incorrectly Classified Time (S)
Instances (%) Instances (%)
Naive Bayes 90.23 12.03 0.05
Random Forest 96.45 0.97 0.02
OneR 55.17 60.33 0.03
SMO 7.14 5.09 0.03
Fig. 5 shows the classification accuracy value of all classifiers.

7
The International Conference on Mathematics & Data Science (ICMDS) 2020 IOP Publishing
Journal of Physics: Conference Series 1743 (2021) 012035 doi:10.1088/1742-6596/1743/1/012035

Figure 5. Accuracy value of all algorithms.

Fig. 6 shows the training time of all classifiers.

Figure 6. Time taken by all algorithms

From the following equations, we give the sensitivity values, precision, specificity, recall, and F-
Measure for all the algorithms. Table. 4 represents the sensitivity, Precision, Specificity, Recall, and F-
Measure Measure Value of all classifiers.

!"
Precision= (2)
!"#$"
!"
Sensitivity= !"#$% (3)
!%
Specificity= !%#$" (4)
!"
Recall= (5)
!"#$%
'∗")*+,-,./∗0*+122
F-Measure = ")*+,-,./#0*+122
(6)

Table 4. Values of Sensitivity, Specificity, Precision, Recall, and F-measure for algorithms.
Algorithms Precision Recall F-measure
Naive Bayes 0.91 0.85 0.83
Random Forest 0.56 0.67 0.56
OneR 0.78 0.58 0.91
SMO 0.89 0.92 0.77
Fig. 7 shows the performance results of each classifier in terms of Precision, Recall, and F-
measure.

8
The International Conference on Mathematics & Data Science (ICMDS) 2020 IOP Publishing
Journal of Physics: Conference Series 1743 (2021) 012035 doi:10.1088/1742-6596/1743/1/012035

Figure 7. Graphical representation of Precision, Recall, and F-measure.

Table. 5 represents the sensitivity, Specificity, and Accuracy value of all classifiers.
Table 5. Value of Sensitivity, Specificity, and Accuracy
Algorithms Sensitivity Specificity Accuracy
Naive Bayes 97.34% 50.34% 85.11%
Random Forest 98.56% 84.50% 96.05%
OneR 92.95% 18.27% 70.40%
SMO 81.30% 77.17% 89.62%

Figure. 8 represent the performance results of Sensitivity, Specificity and Accuracy for each classifier.

9
The International Conference on Mathematics & Data Science (ICMDS) 2020 IOP Publishing
Journal of Physics: Conference Series 1743 (2021) 012035 doi:10.1088/1742-6596/1743/1/012035

In this experimental study, four machine-learning algorithms were used Random Forest, OneR,
SMO, and Naive Bayes. All these algorithms were applied to the diabetic patient dataset. Predicting
accuracy is the main evaluation parameter that we used in this work. From Table. 5, we can notice that
the precision obtained by the Random Forest (98.56%) is better than the precision obtained by SMO,
OneR, and Naive Bayes. It is also easy to see that the Random Forest has the highest value of the
correctly classified instances and the lowest value of the incorrectly classified instances compared to
other classifiers (see Fig. 8).
The accuracy of the algorithms was measured and presented in Fig. 8. OneR gives 70.40% accuracy,
SMO gives 89.62% accuracy, 96.05% accuracy was achieved by Random Forest, and Naive Bayes gives
85.11 %. Therefore, Random Forest achieved the highest accuracy, which is 96.05%. From the
experimental results obtained. Therefore, the study leads us to conclude that the Random Forest
algorithm achieves the best performance in terms of precision and accuracy.

6. Conclusion
In this study, several machine-learning algorithms are applied for classification on a data set. So, in this
study, we have employed four main algorithms: OneR, Random Forest, SMO, and Naive Bayes on the
diabetic datasets. These algorithms have been used for experimentation on WEKA tool to predict
Diabetic patient data. We tried to compare the efficiency and the effectiveness of the cited algorithms
in terms of accuracy, precision and sensitivity. The most objective is to choose the best classification
accuracy. The overall performance of the Random Forest algorithm to predict diabetes disease is better
than SMO, OneR, and Naive Bayes algorithms. In the future work will focus on the integration of other
methods into the used model for tuning the parameters of models for better accuracy.

References

[1] G, J., Buyya. G.R, Marusic. S, &Palaniswami. M., "Internet of Thnigs(IoT) A vision,
architectural elements, and future directions". Future generation computer systeme, 1645-1660,
2013.
[2] Yuehong YIN, Y. Z," The internet of things in healthcare: An overview". Journal of Industrial
Information Integration, 2016.
[3] L. Guariguata, D.R. Whiting, I. Hambleton, J. Beagley, U. Linnenkamp, J.E. Shaw, "Global
estimates of diabetes prevalence for 2013 and projections for 2035", Diabetes Res. Clin. Pract.
vol.103, pp:137–149, 2014.
[4] P. K. Dhillon, S. Kalra, "Secure multi-factor remote user authentication scheme for Internet of
Things environments", International Journal of Communication Systems, 2017.
[5] Rghioui, A.; Lloret, J.; Sendra, S.; Oumnad, A. Glucose Data Classification for Diabetic patient
Monitoring. Applied Sciencs. 2019, 9, 4459.
[6] Z. Mian, K. L.Hermayer, A. Jenkins, "Continuous Glucose Monitoring: Review of an Innovation
in Diabetes Management", The American Journal of the Medical Sciences Vol. 358, pp: 332-339,
Issue 5, November 2019.

10
The International Conference on Mathematics & Data Science (ICMDS) 2020 IOP Publishing
Journal of Physics: Conference Series 1743 (2021) 012035 doi:10.1088/1742-6596/1743/1/012035

[7] A. Rghioui, J. Lloret, M. Harrane, A. Oumnad. “A Smart Glucose Monitoring System For
Diabetic Pateint”. Electronics.9 (4), 678, 2020.

[8] [12] N. M. S. kumar, T. Eswari, P. Sampath, and S. Lavanya,“Predictive Methodology for

Diabetic Data Analysis in Big Data,”ProcediaComput. Sci., vol. 50, pp. 203–208, Jan. 2015

[9] Ahmed, H.B.; Serener, A. Effects of External Factors in CGM Sensor Glucose Concentration
Prediction. Procedia Comput. Sci. 2016, 102, 623–629

[10] K Saravananathan, Velmurugan T, "Analyzing Diabetic Data using Classification Algorithms in

Data Mining", Indian Journal of Science and Technology 9(43), 2016.

[11] Dr Saravanakumar N M, Eswari T, Sampath P and Lavanya S,” Predictive Methodology for
Diabetic Data Analysis in Big Data”, 2nd International Symposium on Big Data and Cloud
Computing,2015.

[12] Dost Muhammad Khan1, Nawaz Mohamudally2, “An Integration of K-means and Decision Tree
(ID3) towards a more Efficient Data Mining Algorithm ”, Journal Of Computing, Volume 3,
Issue 12, December 2011.

[13] Luca Catarinucci, Danilo de Donno, Luca Mainetti, Luciano Tarricone, An IoTAware
Architecture for Smart Healthcare Systems, IEEE Internet Things J. 2 (6) (2015) 515–526.

[14] S, Liver Disease Prediction Using Bayesian Classification, Special Issues, 4th National
Conference on Advance Computing, Application Technologies, May 2014.

[15] B. D. Kanchan; M. M. Kishor. Study of machine learning algorithms for special disease
prediction using principal of component analysis. International Conference on Global Trends in
Signal Processing, Information Computing and Communication. 2016, 5. 5-10.

[16] O. N. Al Sayaydeha ; M. F. Mohammad, “Diagnosis of The Parkinson Disease Using Enhanced

Fuzzy Min-Max Neural Network and OneR Attribute Evaluation Method”, International
Conference on Advanced Science and Engineering (ICOASE), 2019.

[17] F. Wu ; D. Shao, "Study of SMO Parameter Optimization Based on PMSM Finite Element
Simulation", IEEE 3rd Advanced Information Management, Communicates, Electronic and
Automation Control Conference (IMCEC), 2019.

11