See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/371247566

Developed Fall Detection of Elderly Patients in Internet of Healthcare Things

Article in Computers, Materials & Continua · August 2023

DOI: 10.32604/cmc.2023.039084

CITATIONS READS

3 527

3 authors:

Omar Reyad Hazem Ibrahim Shehata

Warsaw University of Technology Shaqra University
53 PUBLICATIONS 703 CITATIONS 11 PUBLICATIONS 101 CITATIONS

SEE PROFILE SEE PROFILE

Mohamed Esmail Karar

Menoufia University
82 PUBLICATIONS 1,952 CITATIONS

SEE PROFILE

All content following this page was uploaded by Mohamed Esmail Karar on 01 September 2023.

The user has requested enhancement of the downloaded file.

 Tech Science Press

DOI: 10.32604/cmc.2023.039084
Article

Developed Fall Detection of Elderly Patients in Internet of Healthcare Things

Omar Reyad1,2 , Hazem Ibrahim Shehata1,3 and Mohamed Esmail Karar1,4, *

1
College of Computing and Information Technology, Shaqra University, P. O. Box 33, Shaqra, 11961, Saudi Arabia
2
Faculty of Computers and Artificial Intelligence, Sohag University, Sohag, 82524, Egypt
3
Faculty of Engineering, Zagazig University, Zagazig, 44519, Egypt
4
Faculty of Electronic Engineering, Menoufia University, Menouf, 32952, Egypt
*Corresponding Author: Mohamed Esmail Karar. Email: mkarar@su.edu.sa
Received: 10 January 2023; Accepted: 26 May 2023; Published: 30 August 2023

Abstract: Falling is among the most harmful events older adults may
encounter. With the continuous growth of the aging population in many
societies, developing effective fall detection mechanisms empowered by
machine learning technologies and easily integrable with existing healthcare
systems becomes essential. This paper presents a new healthcare Internet
of Health Things (IoHT) architecture built around an ensemble machine
learning-based fall detection system (FDS) for older people. Compared
to deep neural networks, the ensemble multi-stage random forest model
allows the extraction of an optimal subset of fall detection features with
minimal hyperparameters. The number of cascaded random forest stages is
automatically optimized. This study uses a public dataset of fall detection
samples called SmartFall to validate the developed fall detection system.
The SmartFall dataset is collected based on the acquired measurements
of the three-axis accelerometer in a smartwatch. Each scenario in this
dataset is classified and labeled as a fall or a non-fall. In comparison to the
three machine learning models—K-nearest neighbors (KNN), decision tree
(DT), and standard random forest (SRF), the proposed ensemble classifier
outperformed the other models and achieved 98.4% accuracy. The developed
healthcare IoHT framework can be realized for detecting fall accidents
of older people by taking security and privacy concerns into account in
future work.

Keywords: Elderly population; fall detection; wireless sensor networks;

Internet of health things; ensemble machine learning

1 Introduction
One of the most widespread health disorders affecting elderly patients is falling. Fall is the second
most frequent global cause of unexpected injuries and fatalities. According to the World Health
Organization (WHO), a count rate of 30% of people over 65 years old suffer accidentally from one
or more falls per year, and this rate increases to match a percentage of 50% for people over 80

This work is licensed under a Creative Commons Attribution 4.0 International License,
which permits unrestricted use, distribution, and reproduction in any medium, provided
the original work is properly cited.
1690 CMC, 2023, vol.76, no.2

years [1]. Thus, the fall required prompt medical assistance. When a fall happens, the fall detection
system alerts the caregivers, minimizing the consequences for the patient. The fall detection system
facilitates the patient’s access to medical care while reducing the negative impacts of falls in realistic
environments [2]. It is one of the primary factors behind the development of many intelligent fall-
detection devices used today. The sensor(s) that gather the environmental and physiological info from
the observed person is vital to any fall detection system. According to Mozaffari et al. [3], fall detection
sensors can be categorized into three groups: ambient (or environmental), motion, and physiological.
The elderly patient’s internal environment is scanned using ambient sensors. The most typical ambient
sensors monitor vision and sound (such as microphones and surveillance cameras). The most popular
motion devices include magnetometers, gyroscopes, and accelerometers [4]. Accelerometers measure
the rate of change in velocity with respect to time, whereas gyroscopes monitor angular velocity in three
dimensions. Finally, magnetometers exist that can recognize orientation. The physiological sensors
analyze critical human body indicators, including blood pressure, oxygen saturation, and temperature.
To evaluate the sensor data and search for specific patterns that identify fall events, fall detection
algorithms use a variety of methodologies [5]. Innovative fall detection models can be used in the
Internet of healthcare things framework based on ensemble machine learning (ML) techniques.
Ensemble machine learning is a technique for obtaining harmony in predictions by combining the key
features of two or more base models. Because ensemble learning minimizes the variation in prediction
errors, the final predictive framework is more stable than the individual models that make up the
ensemble [6]. An aggregated mapping function is produced via ensemble learning, which integrates
the mapping functions encountered by various classifiers. Several approaches presented over the
years employ multiple methods for generating this fusion. The three primary categories are bagging,
stacking, and boosting ensemble learning methods. Bagging includes averaging the predictions from
many decision trees fitted to various samples of the same dataset. When numerous distinct model types
are provided to the same data, stacking is used to learn how to combine the predictions effectively. A
weighted average of the predictions is produced by boosting, which entails adding ensemble members
sequentially that corrects the predictions provided by earlier models [7]. Fig. 1 illustrates the various
elements of the IoHT framework, including mobiles, apps, sensors, devices, and other healthcare
workers.

Figure 1: Schematic diagram of the Internet of health things (IoHT) components

CMC, 2023, vol.76, no.2 1691

In recent years, it has been found that the Internet of medical things (IoMT) and IoHT
technologies can provide the computing capabilities required for fall detection systems [5,8]. Numerous
IoHT-based fall detection methods for older adults have been presented in [9,10]. These IoHT-based
frameworks may be more scalable than conventional structures, making them useful when there
aren’t enough healthcare professionals to handle large numbers of patients. The ambient platform
is interconnected with various IoHT data sources from the health enterprise, including clinical and
administrative applications, sensors, building systems, and medical equipment (and local systems).
Patients can connect to the ambient platform using smartphones, remote devices, and increasingly
embedded devices. These several different data sources will be used by the operations created by the
ambient solution [11].
This study presents a modern IoHT-based fall detection model to help elderly patients’ indoor
healthcare. Following are the contributions summary of this study:
• Developing a framework for a simple and intelligent fall detection model for senior individuals
based on wearable monitors and IoHT benefits.
• Implementing an ensemble-based classifier for the detection of accident falls of elderly patients,
which is suitable for usage in the IoHT environments.
• Conducting a comparison analysis to verify that the proposed ensemble learning model
outperforms conventional machine learning and deep learning (DL) classifiers.
The remainder of this research article is organized as follows. Section 2 presents relevant research
on fall detection systems, including types of sensors, datasets, and machine and deep learning methods.
Section 3 fully describes the proposed IoHT-based framework for fall detection of elderly patients
using the developed ensemble random forest classifier. Overall results and evaluations of the extensive
experiments carried out in this study and the discussions are demonstrated in Sections 4 and 5,
respectively. Concluding remarks and main prospects of this study are finally given in Section 6.

2 Related Works
The evolution of modern technologies developed by researchers to identify and prevent falls in
elderly patients has progressed significantly over the past few years. Several sophisticated techniques
have been presented to tackle the issue of senior people falling [5,12,13]. These solutions include the
use of IoT devices, ML and deep learning algorithms, and imaging training methods. Vision, sound,
or a wearable device can all be used by intelligent fall detection tools to identify falls. Wearable fall
detection sensor nodes mainly use gyroscopes, accelerometers, or a combination of different sensors.
Most tools are monitoring and alarm systems designed to stop, identify and alert care providers to a
fall event.
Additionally, a variety of wearable sensors have been developed in the framework of the IoT in
healthcare to identify and prevent falls in elderly patients when they are at home. More than half of all
hospitalizations for injuries are caused by older patients over 65. As a result, governments are investing
more funds in improving fall detection appliances to decrease the cost of post-fall injury medical care
and treatment [14].
The Whoops system is illustrated with scalable architecture in the work presented in [11]. The
proposed design can track the cases of thousands of older people, catch falls, and inform caregivers.
Additionally, many ML strategies for assessing the applicability of the detection system have been
verified. The recognition accuracy of boosted decision trees (BDTs) was the highest among all the
tested models. Yacchirema et al. in [15] suggested a new IoT-based solution for detecting senior
1692 CMC, 2023, vol.76, no.2

persons’ falls in enclosed locations. This approach integrates wireless sensor networks, smart devices,
and cloud computing platforms. The system develops a new ML model each time a fall is identified
using information from prior falls. With an identification accuracy of 92%, the proposed technique
demonstrates a high level of falling recognition.
Instead of using Red, Green, and Blue (RGB) based cameras for capturing digital images
of elderly persons, an intelligent wearable sensor is adopted, which offers privacy and powerful
light intensity adjustments. To detect falls using an IoT framework, an integrated support vector
machine (SVM) and a histogram of oriented gradients (HOG) have been applied [16]. After acquiring
speckle noise binary images, the HOG algorithm extracts an individual’s attributes. Then, using
linear SVM, these attributes are classified to estimate the falling parameters. The authors employed
long short-term memory (LSTM) model-based edge computing [17] to identify everyday patient
activities, including fall incidents, in real-time. The edge computing platform can detect falls with
an accuracy rate of 95.8% using actual human data stream processing. The study in [18] developed
an intelligent IoTE-fall solution. The IoTE is an IoT platform built on a big data paradigm that
employs ML processing algorithms based on ensemble random forest to detect falls in elderly persons
in indoor surroundings. Three different types of falls—lateral, backward, and forward—were used to
measure the system’s performance. The overall performance metrics for the suggested system—98.72%
recognition accuracy, 96.22% precision, and 94.60% sensitivity—show that it achieves a significant
success rate.
For real-time monitoring of many populations, the authors suggested a centralized IOT-based
fall detection system [19]. Monitoring a massive community can be done using various specialized
devices, such as Arduino and Raspberry Pi boards. Overall, the suggested detection approach scored
a 99.7% accuracy rate. To determine fall events using wearable IoT altimeter sensing devices, the
authors of [20] introduced two temporal inference models, clustering models I and II. Based on these
inferred models, the results for indoor fall monitoring of elderly persons were significant, yielding
the maximum predictions overall accuracy of 98% for the suggested data classification model II. The
authors reported a method for preventing falls in older adults [21] by developing and implementing
a fall monitoring system incorporating ML to make decisions and the IoT to preserve data and send
out alarms. The used ML algorithm has a 96% accuracy rate and is called XGBoost. It is well known
for its speed and accuracy advantages.
The researchers of [22] developed a noise-tolerant FDS that functions well when missing values are
in the data. The study combined recurrent neural networks (RNNs) with an underlying bidirectional
long short-term memory (BiLSTM) structure to build FDS based on wearable sensors. The outcomes
reveal that BiLSTM is a good model for handling missing values in wearable falling detection systems
due to its ability to maintain long-term connectivity. The proposed study in [23] presented an improved
Archimedes optimization method (IAOM) with DL augmented Fall detection model to distinguish
between fall and non-fall events. The IAOM significantly enhances the detection of falls functionality
by using a superior set of CapsNet hyperparameters. A radial basis function (RBF) structure is also
used to identify the relevant class labels for the test images. According to the research results, the
modified IAOM approach has an accurate score of 99.7%. A comparison of current fall detection
approaches is shown in Table 1, depending on the related methodologies used.
CMC, 2023, vol.76, no.2 1693

Table 1: Summary of several fall detection techniques described in the previous studies
Method Types of sensors Dataset Advantages Disadvantages
BDTs [11] Accelerometer & SisFall Small size of stored Unnecessary
Gyroscope and transmitted data emergency
actions
BigML [15] Accelerometer SisFall High success rate in Error rate of 33%
fall detection
HOG-SVM [16] Deep sensor Mixed Accuracy of 98.1% Still detection
errors
LSTM [17] MetaMotionR MobiAct Real-time streaming Data reduction
wearable with 95.8% accuracy
Ensemble- RF [18] 3D-axis SisFall Accuracy of 98.72% Several tri-axial
accelerometer devices
Linear classifier [19] Accelerometer tFall Short response time Need robust
connectivity
Temporal inference Wearable altimeter Synthetic YouTube Falls incident early Methodological
[20] sensor videos warning limitation
XGBoost [21] Accelerometer & SisFall Increased accuracy Young
Gyroscope and reduced false individuals’ data
alarms included
RNN with BiLSTM Accelerometer & SisFall & UP-Fall Handle missing Decrease
[22] Gyroscope detection values in the data incorrect
predictions
CapsNet & Radial Accelerometer UR & Accuracy of 99.7% Unimodal fusion
basis function [23] Multiple model
Cameras Falls

3 Fall Dataset and Methodology

3.1 Fall Dataset
For this work, Texas State University’s publicly available fall dataset was utilized as described in
[24]. That dataset was collected from 14 healthy persons using a wearable smartwatch. They range in
age from 21 to 60 years old, their lengths range from 1.52 to 1.98 meters, and their weights range from
45 to 104 kilograms. The participants generated fall-like scenarios to provide two dataset subclasses
of fall and non-fall cases. The integrated accelerometer’s readings in the x, y, and z coordinates are
included in the entire dataset. The results are expressed in binary form, with ones and zeros marking
instances of falls and non-falls, respectively. A majority of 183806 samples from the fall dataset are
shown in Table 2, partitioned into two main files for the training and testing processes. The total testing
samples are 91025, which includes 5025 fall instances and 86000 non-fall instances. Also, the table
shows 92781 training samples, from which 8175 are for fall cases and 84606 for non-fall cases.
1694 CMC, 2023, vol.76, no.2

Table 2: The fall dataset for this research work is arranged into samples
Data samples Falls Non-falls Overall
Training 8175 84606 92781
Testing 5025 86000 91025
Sum 13200 170606 183806

3.2 Ensemble Machine Learning Model

Random forest (RF) is a decision-tree ensemble technique with fewer hyper-parameters than deep
neural networks [25]. It is developed in a cascade structure such that each level of the cascade receives
feature information processed by the previous level, as depicted in Fig. 2. The levels of ensemble forest
levels are estimated based on the processed input data automatically.

Figure 2: Schematic diagram of proposed ensemble random forest model

One level of the ensemble random forest presents an ensemble of random forests. Each random
forest consists of 500 decision trees, where the number of trees is a hyperparameter for each forest. The
number of input features d is randomly selected to achieve the best value of Gini for tree splitting [26].
Each RF can calculate a class distribution for a given instance by estimating the percentage of different
classes of training samples at the leaf node where the instance is located and then averaging over all
decision trees in the same RF. Each class vector is formed based on the estimated class distribution.
The k-fold cross-validation technique [27,28] has been applied to generate the resulting class vector
for each random forest to avoid overfitting the overall ensemble RF model.

3.3 Proposed Fall Detection System

Fig. 3 depicts the workflow of the developed fall detection system for old patients with smart-
watches in healthcare IoT cloud services. The developed system assumed that the accelerometer sensing
CMC, 2023, vol.76, no.2 1695

device is available in the smartwatch to measure the 3D-axis cartesian position of the patient to
recognize the situation of non-fall and fall-accident situations. The three stages of the developed
healthcare framework to detect fall accidents are described as follows.

Figure 3: Workflow of the IoT-based healthcare framework developed in this study to detect the
situation of fall accidents for older people with a smartwatch to send three-axis accelerometer data
to a cloud server for analysis by proposed ensemble machine learning classifier

Firstly, elderly patients with a smartwatch and three-axis accelerometer readings are used to
measure daily body motion in x-y-z coordinates, as described previously in [29]. The acquired readings
of the senior patient are continuously monitored and sent via the Internet to a healthcare cloud server
to be analyzed by the proposed ensemble machine learning model, as shown in Fig. 3.
Secondly, healthcare cloud services allow the monitoring, analyzing, and archiving of different
forms of patient data, such as reports, signals, and images, without any need for high-cost computing
resources at hospitals, clinics, and medical centers [30]. As shown in Fig. 2, obtained accelerometer
data measurements of the patient motions are automatically classified using the proposed ensemble
RF model.
Finally, the ensemble machine learning classifier gives two output cases: non-fall situation as a
normal behavior and fall accident or emergency case. If a fall accident is detected, it is an emergency.
The healthcare server automatically calls the emergency unit, medical staff (including caregivers and
medical professionals), and family members, as depicted in Fig. 3.
1696 CMC, 2023, vol.76, no.2

3.4 Fall Detection Analysis

The gold-standard evaluation metrics for analyzing fall detection have been used to assess the
proposed ensemble classifier. Cross-validation estimation is used to create a confusion matrix of
2 × 2 [31]. True Positive (TP), True Negative (TN), False Positive (FP), and False Negative (FN) are
the four predicted outcomes of non-fall and fall classes. Additionally, four well-known classification
metrics, namely accuracy, precision, recall (sensitivity), and F1-measure, have been calculated, as given
in Eqs. (1)– (4).
TP + TN
accuracy = 100% (1)
TP + FP + FN + TN
TP
precision = (2)
TP + FP
TP
recall (sensitivity) = (3)
TP + FN
2 (precision × recall)
F1 − measure = (4)
precision + recall

4 Experimental Results and Evaluation

The intended ensemble random forest models were deployed using the Tensorflow 2 and Keras
packages [32]. The classification tests were carried out on a laptop with 16 GB of RAM and an
Intel (R) Core (TM) i7-2.2 GHz processor. All investigations are performed using an NVIDIA 4GB
graphics processing unit (GPU). A total of 92781 and 91025 samples are incorporated in two files
of the public SmartFall dataset [24] for the training and testing processes, respectively. As shown
in Table 2, the testing file contains 86000 patterns of non-fall cases, 17 times more than the 5025
instances of fall cases. All evaluated models’ inputs were taken straight from the actual values of the
accelerometer components (x, y, and z) in the employed dataset without any quantization or other
modifications. In the suggested ensemble random forest classifier, three machine learning models—
K-nearest neighbors (KNN), decision tree, and classic random forest—have been built and tested for
identifying fall situations. This makes it possible to confirm the usefulness of the presented ensemble
random forest model.
Based on the four-classification metrics—accuracy, precision, sensitivity or recall, and F1-score—
presented in Table 3 are the quality assessment results for all evaluated machine learning classifiers
(1–4). The suggested ensemble model successfully fulfilled the optimal classification metrics, yielding
the highest accuracy score of 98.40%, as shown in Fig. 4. In addition, Table 4 shows a comparison
between the suggested ensemble model’s performance and that of other machine learning models from
earlier studies that used the same publicly available dataset. By reaching the greatest accuracy score of
98.40%, the ensemble random forest classifier outperformed the typical deep learning model utilizing
convolutional neural network (CNN) architecture, which had the lowest accuracy value of 86.00%.
With a second-best accuracy score of about 97% for identifying fall occurrences, lightweight CNN can
be used on mobile-embedded devices with constrained computational power. The proposed ensemble
model attains the lowest computational complexity at testing time, and for certain problems of fall
and non-fall classes in real-time scenarios, it can be the best choice.
CMC, 2023, vol.76, no.2 1697

Table 3: Evaluated performance metrics of all tested fall detection models in this study
Fall detection model Class Precision Recall (Sensitivity) F1-score Accuracy (%)
K-nearest neighbors Non-fall 0.958 0.969 0.956 92.90
Fall 0.287 0.229 0.257
Decision tree Non-fall 0.959 0.908 0.928 88.10
Fall 0.158 0.301 0.209
Non-fall 0.948 0.977 0.968 93.80
Random forest Fall 0.346 0.202 0.252
Fall 0.810 0.120 0.221
Proposed ensemble Non-fall 0.982 0.992 0.993 98.40
model Fall 0.831 0.719 0.778

Figure 4: Results of fall detection models accuracy

Table 4: On the same dataset used in this study, comparative of ensemble forest’s performance with
other fall detection models from the literature
Fall detection model Accuracy (%)
Deep learning model [24] 86.00
Long Short-Term Memory (LSTM) [33] 93.46
Lightweight convolutional neural network [34] 96.79
Proposed ensemble model 98.40

5 Discussion
As shown in Tables 3 and 4, the results of this work indicated that the proposed ensemble random
forest classifier is accurate and effective in detecting fall cases in individuals compared to classical
machine learning and deep learning classifiers. The public smartwatch dataset was used to estimate
the suggested classifier’s accuracy score, which is 98.40%. That is fair to assume to be the fundamental
1698 CMC, 2023, vol.76, no.2

module of the proposed medical IoT system, shown in Fig. 3. The following are the principal factors
contributing to the proposed ensemble random forest classifier’s performance success. First, as shown
in Fig. 2, the ensemble forest model’s flow structure enables the generation of new features from the
primary input features. Second, automatic ensemble random forest level or layer number estimate
enables a strong matching to a challenging dataset, such as fall classification data. Finally, compared
to other deep neural models, such as CNN, the number of parameters in the proposed ensemble
classifier is smaller. As a result, the user can easily adjust the ensemble forest model’s attributes. Even
with early stopping constraints, the ensemble random forest classifier’s effective layer estimations
take a long time. However, the provided medical IoT-based design, as seen in Fig. 3, uses cloud
computing services to make computational activities easier to complete. Additionally, the ensemble
forest model’s hyperparameter tuning can be carried out automatically by employing bio-inspired
optimization techniques, including genetic algorithms (GAs) and particle swarm optimization (PSO)
[35]. However, choosing the best set of model hyperparameters requires an optimization technique,
which also takes time.
Recent healthcare IoT edge computing solutions [36] include advanced security and privacy
capabilities for fall detection systems that are not considered in this study or other conventional fall
detection devices [37]. Encryption and decryption operations should also be a part of trustworthy
IoT-based healthcare platforms because they become crucial features for handling real-time tasks.
Upcoming fall detection framework versions could be developed in areas of research like edge IoT
layer [38], explainable and generalizable deep learning [39], and federated learning [40]. However, the
real solution deployment of the suggested medical IoT architecture for fall detection of elderly persons
remains valid and concise.

6 Conclusion and Future Work

Fall detection techniques are crucial for the health and well-being of older persons and, thus,
are essential for their care. An investigation is required to develop solutions to reduce the harmful
impacts of human falls. Such solutions would be more beneficial if they could be tightly integrated
with healthcare management systems [41]. This study provides a straightforward and computationally
effective solution for intelligent fall detection. The designed medical IoT solution makes use of
inexpensive sensors that may be worn and implemented in homes and other institutions. With a
remarkable accuracy score of 98.40%, the presented approach makes good use of an ensemble random
forest algorithm and IoHT devices to identify the occurrence of older people falls in intelligent
environments. For future works, other cutting-edge lightweight DL models for classification could be
employed with different architectures and parameters to enhance the performance of the fall detection
classifiers. Additionally, different types of datasets could be used for training and testing the new
approaches.

Acknowledgement: The authors extend their appreciation to the Deputyship for Research & Innova-
tion, Ministry of Education in Saudi Arabia for funding this research work through the project number
(IFP2021-043).

Funding Statement: This research received the support from the Deputyship for Research & Innova-
tion, Ministry of Education in Saudi Arabia through the project number (IFP2021-043).

Conflicts of Interest: The authors declare that they have no conflicts of interest to report regarding the
present study.
CMC, 2023, vol.76, no.2 1699

References
[1] Falls, “World Health Organization,” Accessed on 30 December 2022. [Online]. Available: https://www.who.
int/en/news-room/fact-sheets/detail/falls
[2] R. Igual, C. Medrano and I. Plaza, “Challenges, issues and trends in fall detection systems,” BioMedical
Engineering Online, vol. 12, no. 66, pp. 1–24, 2013.
[3] N. Mozaffari, J. Rezazadeh, R. Farahbakhsh, S. Yazdani and K. Sandrasegaran, “Practical fall detection
based on IoT technologies: A survey,” Internet of Things, vol. 8, no. 3, pp. 1–40, 2019.
[4] Q. T. Huynh, U. D. Nguyen, L. B. Irazabal, N. Ghassemian and B. Q. Tran, “Optimization of an
accelerometer and gyroscope-based fall detection algorithm,” Journal of Sensors, vol. 2015, pp. 1–8, 2015.
[5] X. Wang, J. Ellul and G. Azzopardi, “Elderly fall detection systems: A literature survey,” Frontiers in
Robotics and AI, vol. 7, no. 71, pp. 1–23, 2020.
[6] R. Polikar, “Ensemble learning,” in Ensemble Machine Learning, 1st ed., vol. 3. Boston, MA, USA:
Springer, pp. 1–34, 2012.
[7] Y. Zhang, J. Liu and W. Shen, “A review of ensemble learning algorithms used in remote sensing
applications,” Applied Sciences, vol. 12, no. 17, pp. 1–20, 2022.
[8] M. E. Karar, H. I. Shehata and O. Reyad, “A survey of IoT-based fall detection for aiding elderly care:
Sensors, methods, challenges and future trends,” Applied Sciences, vol. 12, no. 7, 3276, 2022.
[9] S. Khan, R. Qamar, R. Zaheen, A. R. Al-Ali, A. Al Nabulsi et al., “Internet of things based multisensor
patient fall detection system,” Healthcare Technology Letters, vol. 6, no. 5, pp. 132–137, 2019.
[10] W. Si, R. Tan and G. Yang, “A novel Internet of Things based fall detection system in smart home,”
International Journal of Intelligent Systems, vol. 37, no. 12, pp. 11299–11318, 2022.
[11] D. Mrozek, A. Koczur and B. Małysiak-Mrozek, “Fall detection in older adults with mobile IoT devices
and machine learning in the cloud and on the edge,” Information Sciences, vol. 537, pp. 132–147, 2020.
[12] M. Pech, H. Sauzeon, T. Yebda, J. Benois-Pineau and H. Amieva, “Falls detection and prevention systems
in home care for older adults: Myth or reality?,” JMIR Aging, vol. 4, no. 4, pp. 1–8, 2021.
[13] R. Tanwar, N. Nandal, M. Zamani and A. A. Manaf, “Pathway of trends and technologies in fall detection:
A systematic review,” Healthcare, vol. 10, no. 1, pp. 1–27, 2022.
[14] World Health Organization Europe, “The case for investing in public health,” 2014. [Online]. Available:
https://www.euro.who.int/__data/assets/pdf_file/0009/278073/Case-Investing-Public-Health.pdf
[15] D. Yacchirema, J. Suárez de Puga, C. Palau and M. Esteve, “Fall detection system for elderly people using
IoT and big data,” Procedia Computer Science, vol. 130, pp. 603–610, 2018.
[16] X. Kong, Z. Meng, N. Nojiri, Y. Iwahori, L. Meng et al., “A HOG-SVM based fall detection IoT system
for elderly persons using deep sensor,” Procedia Computer Science, vol. 147, pp. 276–282, 2019.
[17] D. Ajerla, S. Mahfuz and F. Zulkernine, “A real-time patient monitoring framework for fall detection,”
Wireless Communications and Mobile Computing, vol. 2019, pp. 1–13, 2019.
[18] D. Yacchirema, J. S. de Puga, C. Palau and M. Esteve, “Fall detection system for elderly people using IoT
and ensemble machine learning algorithm,” Personal and Ubiquitous Computing, vol. 23, pp. 801–817, 2019.
[19] S. Nooruddin, M. Islam and F. A. Sharna, “An IoT based device-type invariant fall detection system,”
Internet of Things, vol. 9, pp. 1–23, 2020.
[20] G. Kyriakopoulos, S. Ntanos, T. Anagnostopoulos, N. Tsotsolas, I. Salmon et al., “Internet of Things
(IoT)-enabled elderly fall verification, exploiting temporal inference models in smart homes,” International
Journal of Environmental Research and Public Health, vol. 17, no. 2, pp. 1–14, 2020.
[21] D. K. Cahoolessur and B. Rajkumarsingh, “Fall detection system using XGBoost and IoT,” R&D Journal
of the South African Institution of Mechanical Engineering, vol. 36, pp. 8–18, 2020.
[22] M. Waheed, H. Afzal and K. Mehmood, “NT-FDS—A noise tolerant fall detection system using deep
learning on wearable devices,” Sensors, vol. 21, no. 6, pp. 1–26, 2021.
[23] A. S. Alluhaidan, M. Alajmi, F. N. Al-Wesabi, A. M. Hilal, M. A. Hamza et al., “Improved archimedes
optimization algorithm with deep learning empowered fall detection system,” Computers, Materials &
Continua, vol. 72, no. 2, pp. 2713–2727, 2022.
 1700 CMC, 2023, vol.76, no.2

[24] T. R. Mauldin, M. E. Canby, V. Metsis, A. H. H. Ngu and C. C. Rivera, “SmartFall: A smartwatch-based

fall detection system using deep learning,” Sensors, vol. 18, no. 10, pp. 1–19, 2018.
[25] Z. H. Zhou and J. Feng, “Deep forest,” National Science Review, vol. 6, no. 1, pp. 74–86, 2019.
[26] F. T. Liu, K. M. Ting, Y. Yu and Z. -H. Zhou, “Spectrum of variable-random trees,” Journal of Artificial
Intelligence Research, vol. 32, no. 1, pp. 355–384, 2008.
[27] M. E. Karar, A. H. Abdel-Aty, F. Algarni, M. Fadzil Hassan, M. A. Abdou et al., “Smart IoT-based
system for detecting RPW larvae in date palms using mixed depthwise convolutional networks,” Alexandria
Engineering Journal, vol. 61, no. 7, pp. 5309–5319, 2022.
[28] T. T. Wong, “Performance evaluation of classification algorithms by k-fold and leave-one-out cross
validation,” Pattern Recognition, vol. 48, no. 9, pp. 2839–2846, 2015.
[29] O. Ribeiro, L. Gomes and Z. Vale, “IoT-based human fall detection system,” Electronics, vol. 11, no. 4, pp.
1–10, 2022.
[30] M. E. Karar, B. Alotaibi and M. Alotaibi, “Intelligent medical IoT-enabled automated microscopic image
diagnosis of acute blood cancers,” Sensors, vol. 22, no. 6, pp. 1–16, 2022.
[31] M. Sokolova and G. Lapalme, “A systematic analysis of performance measures for classification tasks,”
Information Processing & Management, vol. 45, no. 4, pp. 427–437, 2009.
[32] A. Gulli, A. Kapoor and S. Pal, “Deep learning with TensorFlow 2 and Keras: Regression, ConvNets,
GANs, RNNs, NLP, and more with TensorFlow 2 and the Keras API,” in Expert Insight, 2nd ed.,
Birmingham, UK: Packt Publishing, pp. 1–646, 2019.
[33] M. Farsi, “Application of ensemble RNN deep neural network to the fall detection through IoT environ-
ment,” Alexandria Engineering Journal, vol. 60, no. 1, pp. 199–211, 2021.
[34] S. K. Tasoulis, G. I. Mallis, S. V. Georgakopoulos, A. G. Vrahatis, V. P. Plagianakos et al., “Deep learning
and change detection for fall recognition,” in Engineering Applications of Neural Networks. Cham: Springer,
262–273, 2019.
[35] C. B. Gonçalves, J. R. Souza and H. Fernandes, “CNN architecture optimization using bio-inspired
algorithms for breast cancer detection in infrared images,” Computers in Biology and Medicine, vol. 142,
pp. 105205, 2022.
[36] S. Smys and J. S. Raj, “Future challenges of the Internet of Things in the Health Care domain—An
overview,” Journal of Trends in Computer Science and Smart Technology, vol. 3, no. 4, pp. 274–286, 2021.
[37] A. Tahir, W. Taylor, A. Taha, M. Usman, S. A. Shah et al., “IoT based fall detection system for elderly
healthcare,” in Internet of Things for Human-Centered Design: Application to Elderly Healthcare. Singapore:
Springer, pp. 209–232, 2022.
[38] X. Bai, X. Wang, X. Liu, Q. Liu, J. Song et al., “Explainable deep learning for efficient and robust pattern
recognition: A survey of recent developments,” Pattern Recognition, vol. 120, no. 2, 108102, 2021.
[39] V. Divya and R. L. Sri, “Docker-based intelligent fall detection using Edge-Fog cloud infrastructure,” IEEE
Internet of Things Journal, vol. 8, no. 10, pp. 8133–8144, 2021.
[40] C. Zhang, Y. Xie, H. Bai, B. Yu, W. Li et al., “A survey on federated learning,” Knowledge-Based Systems,
vol. 216, no. 1, 106775, 2021.
[41] A. Mishra, A. Jaiswal, L. Chaudhari and V. Bodade, “Health record management system-A web-based
application,” Journal of IoT in Social, Mobile, Analytics, and Cloud, vol. 3, no. 4, pp. 301–313, 2022.

View publication stats