IoMT based smart healthcare system to control outbreaks of the COVID-19 pandemic

Phase I: infection and disease detection

In the first step, health-related data of COVID patients from various locations is captured and managed using IoT. Early diagnosis of infection is highly needed for the prevention of disease spread, specifically when asymptomatic patients are concerned. This would help in controlling COVID spread as after detection early, patients can be isolated and treated accordingly (Sabeti, 2020). IoT devices can play a significant role in capturing the temperatures as samples for laboratory tests, which makes early diagnosis possible thereby allowing health professionals to devise plans for treatments and save lives. It involves understanding the symptoms like body aches, taste and smell loss, sore throat, cough, vomiting, fever, and diarrhea (Khajenoori et al., 2020). During the infection of COVID, fever is the most common symptom, as it may exceed 100 Fahrenheit (Rahman et al., 2020). Thus, during this phase, IoT can be used for obtaining information.

Phase II quarantine and social distancing

Social distancing and quarantine are needed after the identification of contagious disease to prevent it from being spread with patients still required to be monitored at home or hospitals. Quarantine is recommended for both confirmed and suspected cases. Lockdown which is the sealing of areas of a city or whole country for a certain period for the sake of disease control can be used (Kuhbandner et al., 2020). The main goal is to prevent the disease from being transmitted to patients. In this phase, IoT devices can be used for monitoring patients (Wilder-Smith & Freedman, 2020).

Phase III management & treatment

IoT-based ambulance

During the outbreak of COVID, respiratory tract blockage was the major cause of the increased number of deaths. With such patients requiring intensive care, the medical staff involved with such infected patients remained under high stress. Ambulances based on IoT assisted medical professionals with patients in critical situations (Vistro et al., 2020). For providing fast interaction among actors, a high-speed vehicular network is a must as it allows for attaining the full advantage of these ambulances. Medical professionals access the information of patients remotely by using IoT. Figure 2 shows an image of an IoT-based ambulance.

Social media

Because social media is being used by millions of people, it makes it suitable for connecting people regarding their health. To deal with the COVID pandemic, patients can communicate with physicians and can also attend their live sessions. Resultantly, the number of the hospital for treatment is reduced as there is no need for physical contact for health professionals for giving medical prescriptions to patients (Machado et al., 2020).

Phase IV eradication

Implementations of lockdown whether it is citywide or countrywide has caused economic problems and in the current phase of recovery, crucial actions are being taken by governments for its eradication to the full extent. As institutions and businesses are re-opening in phases, a significant role is also played by IoT for the complete removal of viruses.

Automatic sanitizing spray

To protect people against COVID infection, an epidemic smart tunnel that made use of an automatic sanitizing spray system was proposed (Pandya, Sur & Kotecha, 2020). The proposed system works by detecting humans and their motion thereby disinfectant spraying whenever someone passes through it. Figure 3 shows a sanitizing door designed by a Pakistani electronic corporation Sandoir Technologies.

Phase V preventive measures

Government-provided preventive measures for the people are listed as follows. These measures vary from country to country and city to city within a country concerning the virus’s spread.

1. Use of sanitizers and washing hands with soaps in case of contact with people,

2. Maintenance of social distancing,

3. Pregnant women and old people must stay at home,

4. Touching eyes, ears, mouth, and nose unnecessarily, should be avoided,

5. Floors and handles of doors must be sterilized or sanitized regularly.

Smart city

Due to the COVID outbreak, IoT devices for monitoring the important places of the city like hospitals, airports, bus terminals, markets, and subways are installed. In cities where data related to arrival and departure time is being monitored and analyzed, this concept of a smart city is used. Li (Li, Batty & Goodchild, 2020) studied various such sensors which are installed for providing real-time information at different places. For activity monitoring, authorities are using wearable IoT devices. Figure 4 shows an example smart city.

Scalability

As IoT and devices based on IoT began to grow considerably along with the advancement of technology, scalability is a major challenge. Alongside their use in houses, these devices are used for preventing COVID spread. For tracing the patients of COVID and assessment of social distancing implementation, BLE wearables are used, with their data management, and stored on the cloud and so there is a need for powerful computational systems, and large storage is required. Also with scalability, energy needs are heightened.

Reduced bandwidth and erroneous data

With the outbreak of COVID, the use of IoT-based devices is growing. There are many sensors in IoT-based wearables for controlling the spread of COVID with each sensor sending information by using APIs. Cellular networks are providing bandwidths for the transfer of data between sensors of IoT and clouds but after increased usage of wearable IoT devices, the offered bandwidth is inadequate for transferring real-time data transfer. It sometimes results in the transfer of erroneous data to the cloud, which disturbs the working and efficiency of wearable IoT.

Miscommunication & privacy issues of data

Considering the previously reviewed bandwidth and scalability issues for wearable IoT devices, they are causing other issues regarding privacy and data miscommunication. The consequences of these problems are

1. There is the live transmission of data along with personal information from IoT-based devices which are attached to people. As there is a stability problem in IoT, for the secure transmission of data between the device and the cloud, there are no cryptographic algorithms available. So, data interception on communication channels is possible resulting in leakage of personal and other sensitive information (Kamal & Tariq, 2018; Aman, Basheer & Sikdar, 2019).

2. The access to the personal data of the patients should be prioritized and provided only to the relevant persons.

Interoperability issues

Interoperability is a big concern for the proper functioning of IoMT, after bandwidth, privacy, and scalability issues. For accurately transmitting real-time data, a dynamic scheme is needed. Health officials should be provided training sessions in this regard. During COVID, additional goals should be set for underdeveloped and remote areas (Gupta, Christie & Manjula, 2017).

Collection of data for symptoms of disease

This module uses IoT-based sensors for collecting COVID symptoms including fever, coughing, sore throat, tiredness, and shortness of breath in real-time. Various biosensors are available in the market, for example for the detection of cough, audio sensors with aerodynamics, fever detection temperature sensors, and oxygen-based sensors can be used for shortness of breath. Similarly, images can be used for sore throat detection and chest pain and fatigue can be detected with heart rate sensors.

Quarantine and heath centers

The goal of this component is to keep track of the information on quarantined or isolated patients by recording technical and non-technical data. Non-technical data includes last week’s information like travel and illness history, gender, and age while technical data contains symptoms and the patient’s response in the course of treatment. During treatment, both records are vital to deal with.

Data warehouse

Machine learning algorithms are used for analyzing the data which is uploaded by sensors to data centers and deliver helpful information on real-time data processing. Detection and treatment of COVID are achievable with the help of these models. Researchers can understand the nature of the disease by comprehensive analysis of these results.

Health professionals

Health professionals make use of results generated by machine learning models for dealing with cases of COVID and suspected patients getting quick responses and being screened for confirmation which results in identified patients getting treatment on time. The Internet is used for connecting components of the framework. Sensors upload real-time data of the patients to the servers. After the maintenance of records, results generated by the machine learning models are discussed with physicians and health officials. Additionally, to stop the further spread of the virus, and provide precautions to the public, the proposed framework can also be used.

Data set

A dataset named “COVID Symptoms Checker” is available (https://www.kaggle.com/iamhungundji/covid19-symptoms-checker) and contains features for the classification of whether a person is infected or not. These features are based on indications set by WHO. There are twenty-seven attributes like age, country, symptoms, contact, and severity in the dataset further categorized into five groups. Table 1 details of attributes. In this research, 26 attributes are used for the prediction of the severity level of persons infected with COVID Fig. 6. There are 316,800 records in the dataset and there are 79,200 records having severity associated with it.

Data preprocessing

There are many attributes in the dataset with unique values, like “none”, “mild”, “moderate”, and “severe” which are possible values for severity attributes. There are binary values in “Yes” and “No” format in tiredness, dry cough, fever, sore throat, and difficulty in breathing attributes. Since vector numeric form is required for learning classifiers, so label encoding is applied for each attribute for conversion of text values into respective numeric.

Predictive models

The primary idea of the machine learning models is to assess the present condition of the patient. For assessment and prediction of the severity of disease level in patients infected with coronavirus, a particular dataset is used. The models used in this study are implemented in Python and Scikit-learn (Pedregosa et al., 2011) and the performance of these models is compared.

Random forest (RF) makes use of decision trees and generates trees for reducing the variance. It is equally good for classification and regression problems. The bagging technique is used to make the final prediction using the majority voting from tree results. To escape variance and overfitting, a random subset of features and dataset is used (Breiman, 2001).

AdaBoost (AB) is a boosting procedure and provides outstanding performance for classification problems. It learns from misclassified instances appearing in previous iterations and then transforms the weak learners into strong learners based on its results (Schapire & Singer, 1999). The main idea here is to get better performance with high accuracy by combining with weak learners. AB makes use of feature scores for finding the important features.

Gradient boosting machine (GBM) is a collaborative technique based on boosting and is used in classification and regression. This technique combines weak learners like decision trees for final prediction (Natekin & Knoll, 2013). Boosting model converts weak learners into strong learners with each iteration. For performance improvement, the loss function is used which helps in finding the effectiveness of the coefficients of models over underlying data.

For generating and aggregating results from multiple decision trees, the extra tree (ET) model is used, which makes final predictions. It does so by averaging the decision with various samples of the dataset (Geurts, Ernst & Wehenkel, 2006). Rather than picking the best node, an extremely randomized tree utilizes the whole sample from the dataset and picks the random root node. It is also considered an extremely randomized tree classifier for this reason.

The voting classifier (VC) model uses the probability scores from various base learners and predicts the output. In these types of models, the output can be acquired using two ways

1. Soft voting

2. Hard voting

In soft voting, an average of every model is calculated and then the output value with high probability is chosen whereas, in hard voting, the majority of votes from classifiers are used for the final output. In the current study, soft voting is used and SGD and LR are combined for VC. LR is a linear method that utilizes probabilities for making predictions. For modeling dependent variables, it uses a logistic function (Wright, 1995). It can also be used for classification and regression. Taking derivatives from the training date is gradient descent, while SGD is used for taking derivatives from every entry of training data and was introduced as a technique for variance reduction (Johnson & Zhang, 2013). For classification tasks, SGD and LR are well-known techniques.

Proposed methodology

In state-of-the-art research works, ensemble learning techniques are considered to get more reliable results than individual models. Ensemble learning benefits from the wisdom of the crowd, leveraging the collective knowledge and predictions of multiple models to improve accuracy, robustness, and generalization. These are the reasons that proposed ensemble learning models predict the severity of COVID patients more accurately.

The workflow for predicting the severity of COVID patients is depicted in Fig. 7. The model is a combination of two machine learning models, RF and an extra tree classifier (ETC). Experiments are conducted using the COVID severity prediction dataset. The proposed model is applied to the dataset. The data is split with a ratio of 0.7 to 0.3, where 70% data is used for training and 30% for testing. The performance of the model is evaluated using precision, recall, and F-score metrics.

In this work, ETC and RF models are combined using the soft voting criteria. The final output is decided by using the probability of each class. Algorithm 1 shows how the proposed ensemble model works.

Mathematically, the soft voting criteria can be represented as (1)${\displaystyle \widehat{p}=argmax\sum _i^{n}R{F}_i,\sum _i^{n}ET{C}_i}$

where the probability values against the test sample are denoted by ${\sum }_i^{n}R{F}_i$ and ${\sum }_i^{n}ET{C}_i$ and $\widehat{p}$ shows the final prediction. The probability values for each instance using ETC and RF are then passed through based on soft voting.

Each test sample is fed into ETC and RF models for label prediction. These models assign a probability score to each class of the given sample. For example, if the RF model’s probability values are 0.5, 0.6, 0.7, and 0.8 for four classes, respectively, and the ETC model’s probability values are 0.6, 0.7, 0.8, and 0.9 for four classes, respectively, and P(x) represents the probability value of x ranging from 0 to 1, the final probability is determined as

P(1) = (0.5 + 0.6)/2 = 0.55

P(2) = (0.6 + 0.7)/2 = 0.65

P(3) = (0.7 + 0.8)/2 = 0.75

P(4) = (0.8 + 0.9)/2 = 0.85

The final prediction will be 4 because it has the highest probability. VC(RF+ETC) chooses the final class based on a class’s maximum average probability and combines the predicted probabilities of both classifiers.

Comparison with state-of-the-art techniques

The performance of the proposed model is compared to that of classifiers employed in Hassan, Rashid & Hamarashid (2021) and Rochmawati et al. (2020) applied to the COVID symptoms severity dataset. Table 3 presents the results of all models, which reveals that the Voting classifier has shown superior performance. In Hassan, Rashid & Hamarashid (2021) iECA has shown the highest accuracy of 0.909. Saengamnatdej, Molee & Warnnissorn (2023) observe the effect of SMOTE data upsampling with extreme gradient boosting (XGB) model to predict COVID-19 with an accuracy of 0.812. The research work (Giotta et al., 2022; Mahesh et al., 2022) implemented tree base learning models and get an accuracy of 0.759 and 0.825 respectively. On the other hand, the voting classifier of the current study has shown a 0.922 Accuracy score.

Importance of using IoT

IoT-based devices are and will be contributing significantly to managing the COVID pandemic. These devices do so by real-time monitoring of patients and their data for the sake of locating and identifying the patients. It can also detect fraudulent information from its sources.

Whenever the spread of a pandemic or epidemic is concerned, early identification and tracing of infected people and their isolation are crucial steps. The use of IoT-based devices as a solution for these challenges is getting wide attention. This research highlights the uses of wearable IoT-based solutions for fighting the COVID outbreak. The main goal is tracing the infected persons, spread elimination, and reducing the large impact of the disease.

Using IoT properly will surely aid in the early identification of infected people and making decisions accordingly. IoT-based tools can be applied for capturing data, predicting situations, and enabling government and health officials for making appropriate policies during the pandemic situation.