Analysis of COVID-19 severity from the perspective of coagulation index using evolutionary machine learning with enhanced brain storm optimization

Abstract

Coronavirus 2019 (COVID-19) is an extreme acute respiratory syndrome. Early diagnosis and accurate assessment of COVID-19 are not available, resulting in ineffective therapeutic therapy. This study designs an effective intelligence framework to early recognition and discrimination of COVID-19 severity from the perspective of coagulation indexes. The framework is proposed by integrating an enhanced new stochastic optimizer, a brain storm optimizing algorithm (EBSO), with an evolutionary machine learning algorithm called EBSO-SVM. Fast convergence and low risk of the local stagnant can be guaranteed for EBSO with added by Harris hawks optimization (HHO), and its property is verified on 23 benchmarks. Then, the EBSO is utilized to perform parameter optimization and feature selection simultaneously for support vector machine (SVM), and the presented EBSO-SVM early recognition and discrimination of COVID-19 severity in terms of coagulation indexes using COVID-19 clinical data. The classification performance of the EBSO-SVM is very promising, reaching 91.9195% accuracy, 90.529% Matthews correlation coefficient, 90.9912% Sensitivity and 88.5705% Specificity on COVID-19. Compared with other existing state-of-the-art methods, the EBSO-SVM in this paper still shows obvious advantages in multiple metrics. The statistical results demonstrate that the proposed EBSO-SVM shows predictive properties for all metrics and higher stability, which can be treated as a computer-aided technique for analysis of COVID-19 severity from the perspective of coagulation.

1. Introduction

In December 2019, a serious coronavirus first emerged in Wuhan, China, and later spread rapidly across the world in a short period of time (Li et al., 2020, Hu et al., 2021, Li et al., 2021, Singh et al., 2020). On January 30, 2020, the World Health Organization (WHO) declared that the novel coronavirus outbreaks a Public Health Emergency of International Concern (PHIC) (World Health Organization, 2020). According to the recommendations by the WHO, the novel coronavirus was named coronavirus disease (COVID-19) in February 2020 (World Health Organization, 2020). As of May 23, 2020, there have been a total of 5,103,006 confirmed cases of COVID-19 worldwide with 333401 fatal cases (World Health Organization, 2020). Despite governments worldwide having put great effortsthere is still a lack of effective strategies to control the spread of the disease. The rising number of COVID-19 patients can lead to a shortage of health care resources and increase the workload of medical staff. Once the medical capacity of the hospital is overloaded, it may increase the risk of COVID-19 patient mortality. A recent study has found that if the novel coronavirus is not effectively controlled, it could collapse the entire health care system in a short time (Pourhomayoun and Shakibi, 2020). Therefore, it is urgently required to develop a new prediction model for early recognition and classify the severity of COVID-19. Developing an effective prediction model can help medical institutions decide whether the patient needs to get attention first, whether the patient has a higher priority for hospitalization, whether the patient can get a greater level of monitoring and treatment, and thereby rationally and effectively allocate medical resources and reduce the healthcare burden.

In recent years, machine learning and optimization approaches (or a hybrid of them) have become highly critical in the field of medicine (Jordan and Mitchell, 2015, Rajkomar et al., 2019). Machine learning methods for diagnosing different diseases were widely used, predictive model creation helped clinical decision making, and significant risk factors related to the disease were identified (Obermeyer and Emanuel, 2016, Bhagyashree et al., 2018). Optimization is a root idea in not only development of e-healthcare systems but also areas such as fault diagnosis of rolling bearings (Deng, 2020, Zhao et al., 2019), scheduling problems (Pang et al., 2018, Zhou et al., 2018), bankruptcy prediction (Cai et al., 2019, Yu et al., 2021, Zhang et al., 2020), wind speed prediction (Chen et al., 2019), image segmentation (Zhao et al., 2020, Zhao et al., 2020), engineering design problems (Ba et al., 2020, Gupta et al., 2019, Liang et al., 2020, Zhang et al., 2020). The exploration and exploitation competencies of these stochastic search methods can play a significant role in dealing with medical data classification (Lufeng et al., 2017, Huang et al., 2019, Li et al., 2018, Zhao et al., 2019), PID optimization control (Zeng et al., 2015, Zeng et al., 2014, Zeng et al., 2019), feature selection (Hu et al., 2021, Li et al., 2017, Liu et al., 2015, Zhang et al., 2020, Zhang et al., 2020), hard maximum satisfiability problem (Zeng et al., 2011, Zeng et al., 2012), parameter optimization (Heidari et al., 2019, Shen et al., 2016, Wang and Chen, 2020, Wang et al., 2017), gate resource allocation (Deng et al., 2020, Deng et al., 2020), detection of foreign fiber in cotton (Zhao et al., 2015, Zhao et al., 2014), and prediction cases in educational concerns (Lin et al., 2019, Jixia et al., 2019, Wei et al., 2020, Wei et al., 2017, Zhu et al., 2020). In particular, the machine learning technique may contribute to improving the diagnostic quality, reducing the workload of radiologists, anatomical pathologists, and the accuracy of diagnoses in diseases (Obermeyer and Emanuel, 2016), to minimize error and missed diagnosis incidence. Machine learning techniques have now become an important tool in clinical work. During the COVID-19 outbreak, machine learning methods were developed and used to identify, track and manage COVID-19, COVID-19 epidemic. In recent years, the use of artificial intelligence (AI) is popular in many fields of life sciences (De-Kuang Hwang et al., 2019, Alam et al., 2019). In the field of ophthalmology, for example, AI has achieved the level of a specialist in the distinction of diseases (Kermany et al., 2018). AI can help radiologists to diagnose benign and malignant thyroid nodules in radiology in qualitative terms (Wang et al., 2019). With the rapid growth of AI, AI machine learning technologies have been widely used for disease diagnosis; predictive models have been created to support clinical decision making in medical fields, and key factors related to illness have been quickly established (Obermeyer and Emanuel, 2016, Bhagyashree et al., 2018, Lee et al., 2014). As a result, AI technology based on machine learning is increasingly important in the medical field in computing technology. Machine-based AI technology has also been used for diagnosis of disease (Albahri et al., 2020). Computer tomography (CT) or X-ray image recognition (Kang et al., 2020, Albahli, 2021), disease epidemic, surveillance,and control (Yang et al., 2020, Zheng et al., 2020) in the course of the COVID-19 outbreak. By using classical and ensemble machine learning algorithms, Khanday et al. (2020) categorized textual clinical reports in four groups. Effect design has been carried out using frequency term/inverse document frequency (TF/IDF), word bag, and report length. Term frequency/inverse document length was used. These features have been given to classifying conventional and ensemble machines. Booth et al. (2021) designed a prognostic model for mortality in COVID-19 infection using machine learning. Prakash et al. (2020) used machine learning algorithms to analysis, prediction and evaluation of covid-19 datasets. Erraissi and Banane (2020) used a machine learning model to predict the number of cases contaminated by COVID-19, and the results showed the possibility of achieving better scores prediction after using the proposed method. Many documents show that machine learning algorithms have shown great potential in solving new coronary covid-19-related problems.

This study is first developed from the point of view of the coagulation indexes, an evolutionary SVM to diagnose COVID-19. However, in SVM, penalty factor C and the $\gamma$ value can change the accuracy and efficiency of the learning process, and it is sensitive to these values. Both core operators are employing this proposed method (EBSO-SVM) to improve and re-establish the search capacity for the brain storm optimization algorithm (BSO) (Shi, 2011), abstracted from the Harris hawks optimization, that can ensure substantial convergence and spring potential of the local stagnant. Though, a large number of evolutionary algorithms has been designed such as monarch butterfly optimization (MBO) (Wang et al., 2019), slime mould algorithm (SMA) (Li et al., 2020), moth search algorithm (MSA) (Wang, 2018) and Harris hawks optimization (HHO) (Algorithm and applications, 2019). Typical BSO collects a group of specialists with various backgrounds, expertise, and skills to find a solution to the problem. Due to the effective learning and utilization of the samples sampled in the optimization process and its ease of implementation. The proposed EBSO algorithm is verified on 23 benchmarks, composed of 7 unimodal benchmark functions, six multimodal benchmark functions, and ten fixed-dimension multimodal benchmark functions. Finally, COVID-19 clinical data were used for EBSO-SVM and other SVM competitors based on other optimization algorithms, along with coagulation indexes. The EBSO core compensation is validated by evaluating the experimental findings, and different performance evaluation indexes can establish a strong EBSO-SVM for determining COVID-19 status from the perspective of the coagulation index. The results of the test showed that the EBSO-SVM proposed was seemingly beneficial.

The key contributions in this study are as follows:

• •The two core operators have taken the opportunity to improve the brain storm optimizing algorithm, EBSO, from harris hawks optimization.
• •The proposed EBSO algorithm is verified on 23 benchmarks composed of 7 unimodal benchmark functions, six multimodal benchmark functions, and ten fixed-dimension multimodal benchmark functions.
• •The EBSO has successfully solved the optimization of SVM’S parameters and feature selection at the same time.
• •An efficient EBSO-SVM technique is used to assist COVID-19 diagnosis from the coagulation index perspective.

The paper was organized according to next arrange. The materials and processes are reported in paragraph 2. The proposed EBSO algorithm is defined in Section 3. Section 4 describes the proposed EBSO-SVM model. Section 5 describes the designs of the tests. The results of EBSO-SVM on COVID-19 data set-simulations are described in Section 6, and the results of EBSO on 23 benchmarks are also exhibited in this section. The findings are discussed in Section 7. Section 8 shows the conclusion and path of the future.

2. Materials and methods

2.1. Data collection

The participants in our study were from the Wenzhou Rural retrospective study. The Ethics Committee of the Affiliated Yueqing Hospital of Wenzhou Medical University (approval No. 202000002) approved the present study. A total of 51 participants aged between 18 to 93 years were recruited from the Affiliated Yueqing Hospital of Wenzhou Medical University over the period between January 21 and March 20, 2020. The following general clinical information was collected for each participant: gender, age, and coagulation function indexes. Coagulation functions were measured using an automated blood coagulation analyzer (Sysmex CA-7000 analyzer, Kobe, Japan).

In our study, the diagnosis of COVID-19 was carried out and guided according to the criteria issued by the National Health Commission of the People’s Republic of China. The etiological criteria for diagnosing COVID-19 are at least one of the following: (i) severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) confirmed by positive real-time reverse transcriptase-polymerase chain reaction (RT-PCR) on the nasopharyngeal swab, sputum, or stool samples; (ii) to sequence the virus genome, which is highly homologous to the genome of SARS-CoV-2. According to clinical characteristics, COVID-19 patients are classified into four categories: mild, general, severe, and critically ill patients. Mild patients with the following characteristics: (i) patient has no symptoms or mild symptoms; (ii) patient has no lung involvement. General patients with the following characteristics: (i) the clinical manifestations of COVID-19 patients were fever, dry cough, fatigue, nose congestion, runny nose, myalgia, sore throat, diarrhea, and so on; (ii) the lungs affected by SARS-CoV-2. Severe patients with at least one of the following characteristics: (i) patient present with dyspnea and respiratory rate (RR) greater than 30 breaths/min; (ii) patient blood oxygen saturation lower than $⩽93\%$; (iii) patient oxygenation index (arterial partial pressure of oxygen to fraction inspired oxygen) below 300 mmHg (1 mmHg = 0.133 kPa). Critical ill patients with at least one of the following characteristics: (i) patient required intubation and mechanical ventilation because of acute respiratory failure; (ii) patient present with shock; (iii) patient present with multiple organ failure. On admission, initial COVID-19 severity was assessed by sixth revised trial version of the Novel Coronavirus Pneumonia Diagnosis and Treatment Guidance (El-Solh et al., 2020, Cheung et al., 2021). Cases meeting any of the following criteria were defined as severe cases: (1) RR greater than 30 breaths/min; (2) patient blood oxygen saturation lower than 93%; (3) patient oxygenation index below 300 mmHg; (4) patients with respiratory failure needing mechanical ventilation, septic shock, and/or multiple organ dysfunction. Taken together, depending on the clinical symptoms and clinical guidelines, the confirmed COVID-19 cases were further classified into two categories: mild and general cases were categorized as the non-severe group, severe and critical cases were categorized as the severe group (Szklanna et al., 2021, Suvarna et al., 2021).

2.1.1. Statistical analysis

Statistical analysis was performed with SPSS, version 21 (IBM, Somers, NY, USA). The differences in coagulation function indicators and age between the non-severe COVID-19 and severe COVID-19 groups were analyzed by an independent sample t-test. The data are expressed as the mean $\pm$standard deviation ($\overline{x}$ $\pm$SD). P-values less than 0.05 (P  < 0.05) were considered significant. A total of 9 parameters were used as illustrated in Table 1 . A detailed description of statistical analyses is described in Table 2 .

Table 1.

List of the features used in this study and their definitions.

No.	Feature	Abbreviation
F1	Gender	Gender
F2	Age	Age
F3	Prothrombin time	PT
F4	International normalized ratio	INR
F5	Prothrombin time activity	PTA
F6	Fibrinogen	FIB
F7	Activated partial thromboplastin time	APTT
F8	Thrombin time	TT
F9	D-dimer	D-D

Table 2.

Coagulation function indicators clinical parameter in severe group and non-severe group.

Index	Non-severe (n  = 30)	Severe (n  = 21)	p-value
Age(years)	42.300 ± 11.530	61.43 ± 17.64	0.000
PT (s)	13.030 ± 0.780	14.930 ± 3.050	0.011
INR	1.030 ± 0.060	1.090 ± 0.070	0.003
PTA (%)	95.880 ± 7.350	88.280 ± 8.400	0.001
FIB (g/l)	4.170 ± 0.870	4.970 ± 1.650	0.053
APTT (s)	38.480 ± 4.310	43.650 ± 8.780	0.019
TT (s)	16.210 ± 0.740	16.520 ± 0.820	0.171
D-D(mg/l)	0.480 ± 0.260	1.990 ± 3.750	0.032

2.2. Brief introduction of support vector machine (SVM)

The SVM is the most frequently utilized machine learning model developed based on some mathematical concepts on risk prevention and the idea of the VC dimension. In this model, we intend that reach an excellent compromise between decreasing the error of the output signals in the training domain and optimizing the margin to attain the best generalization capacity and prevent the overfitting concern. In some cases, the SVM was used (Shen et al., 2016, Chen et al., 2011, Chen et al., 2014) to deal with some small sample data sets in particular because of it adequate velocity rate and good classification accuracy.

When we set a suitable w and b for tests, the SVM model is able to distinguish the undecided samples according to the utilization of the hyperplanes with acceptable classification rates. SVM can also tackle non-linear classification problems if we apply kernel-based methods during the modeling. The non-linear feature of concern is modeled as follows:

$$g(x)=\mathit{sgn}({\displaystyle \sum _{i=1}^n}{a}_i{y}_{i}K(x^i,x)+b))$$ (1)

where $K(x,x^i)$ is the kernel function and $K(x,x^i)$ is Gaussian kernel. For more details, readers refer to work by Shen et al. (2016). In SVM, penalty factor C and the $\gamma$ parameter of the kernel can change the accuracy and efficiency of the learning process, and it is sensitive to these values. C can show us how much the SVM can generalize. $\gamma$ can show us how much the model is fitting.

2.3. Brain storm optimization (BSO)

The brain storm optimization (BSO) algorithm has been appeared in 2011 (Shi, 2011). That is a promising and early swarm intelligence algorithm. Moreover, it is focused on the human being’s collective actions, that is, brainstorming. Speciation is a natural selection process, which means the population is differentiated into individual species. BSO solutions also diverge into many clusters. The new solutions are based on the mutation of one particle or two. The basic BSO algorithm is conceptually simple and easy to be programmed. The original algorithm consists of four steps: Initialization, Clustering, New particles’ generation, and Selection. Randomly generate n potential solutions (particles), and evaluate the n particles, Cluster n particles into m clusters by a clustering model, and then randomly select one or two cluster(s) to generate new particle, finally execute the selection operator. The detailed procedure of the BSO can be seen in Algorithm 1. The n is the population size, T is the maximum number of iteration, $X_b$ is the current best position, $f_{\mathit{value}}$ is current best fitness value, $X_i(i=1,2,\dots ,N)$ means the population, $p_1$ and $p_2$ are pre-determined probability.

Algorithm 1: The pseudo-code of original BSO

2.4. Core operators to be introduced

By imitating the cooperation conduct of hawks to catch the prey, the Harris Hawks Optimizer (HHO) is built (Algorithm and applications, 2019). The hawks are one of nature’s cleverest birds. First, the Hawks seek in a number of neighborhood sites for rabbits (bearers) and then hunt rabbits (victim). The rabbits also demonstrate escape behavior by pursuing different methods, thereby increasing their survival chances. The typical HHO imitates these chasing-escaping actions and reactions in order to identify the best answers to a single objective problem of optimization. The main part of HHO is composed of exploration phase, transition phase and exploitation phase. For exploring the search area, the search algorithms are mathematically represented as follows:

$$X_{t+1}=\left\{\begin{array}{cc}{X}_{\mathit{rand},t}-r_1\left|X_{\mathit{rand},t}-2r_2{X}_t\right|\hfill & r_5⩾0.5\hfill \\ (X_{\mathit{rabbit},t}-X_{m,t})-r_3(\mathit{lb}+r_4(\mathit{ub}-\mathit{lb}))\hfill & r_5<0.5\hfill \end{array}\right)$$ (2)

where $X_t{\mathit{andX}}_{t+1}$ are the search agent vector of $\mathit{tandt}+1$ respectively, $X_{\mathit{rabbit},t}$ is location of target preyrabbit, $r_1$, $r_2$, $r_3$, $r_4$, and $r_5$ are all interval of random values (0,1), $\mathit{lbandub}$ the lower and the upper limits of choice $X$ variables, $X_{\mathit{rand},t}$ represents the selected swarm hawk’s random position, and $X_{m,t}$ is the mean swarm state for the $t$ iteration and mean hawk condition is determined as $X_{m,t}=\frac{1}{N}{\sum }_{i=1}^N{X}_{i,t}$. The rabbit’s energy is utilised to make the transformation as $E_R=2E_0\left(1-\frac{t}{T}\right)$, where $E_R$ symbolizes the rabbit’s escape energy, $E_0$ is the initial state of its strength and $T$ is the maximum number of iterations utilized for the traditional HHO stop criteria, which as transition phase.

The exploitation phase is composed of soft besiege with progressive rapid dives. It is supposed that hawks may evaluate their next step with the following formula, to conduct a soft besiege:

$$Y=X_{\mathit{rabbit},t}-E_R\left|{\mathit{JX}}_{\mathit{rabbit},t}-X_t\right|$$ (3)

$$Z=Y+R\times L_F(D)$$ (4)

where $D$ shows the problem dimension, and $R$ represents a random $1\times D$ size vector and $L_F$ is the random vector distributed by levy-flight and can be determined as follows:

$$L_F(y)=\frac{u_1\times \sigma }{{\left|u_2\right|}^{\frac{1}{\alpha }}},\sigma ={\left(\frac{\mathrm{\Gamma }(1+\alpha )\times \mathit{sin}(\frac{\pi \alpha }{2})}{\mathrm{\Gamma }(\frac{1+\alpha }{2})\times \alpha \times 2^{(\frac{\alpha -1}{2})})}\right)}^{\frac{1}{\alpha }}$$ (5)

where $u_1$, $u_2$ means random interval values (0,1), $\alpha$ is fixed as 1.5 as constant. Hard besiege with PRD and this strategy models the mechanism utilized as follows:

$$X_{t+1}=\left\{\begin{array}{cc}Y\hfill & \mathit{ifF}(Y)<F(X_t)\hfill \\ Z\hfill & \mathit{ifF}(Z)<F(X_t)\hfill \end{array}\right)$$ (6)

where $\mathit{YandZ}$ vectors can be identified by Eqs. (2), (3).

3. Proposed EBSO

In this study, the two core operators from harris hawks optimization to further enhance and improve the search ability of the original BSO, EBSO is designed. To the best of our knowledge, this is the first time the BSO has been effectively integrated with several efficient operators from harris hawks optimization. The designed algorithm EBSO can be divided into two parts. The first part is to execute each step of the original BSO; the second is to complete the execution of the introduced operators derived from harris hawks optimization. The detailed pseudo-code of the presented EBSO can refer to Algorithm 2.

Algorithm 2: The pseudocode of designed EBSO

4. Proposed EBSO-SVM model

In this study, an evolutionary support vector machine-driven enhanced brain storm optimization algorithm for analysis of COVID-19 from the perspective of coagulation index is designed (EBSO-SVM). The framework of the EBSO-SVM is exhibited in Fig. 1 . From this Figure, we see that the Cross-validation mechanism is utilized as the internal 5- and external 10-fold process, respectively, which means the $K_1=5$ and $K_2=10$ in this paper. The detailed data segmentation diagram is shown in Fig. 2. It shows us that it is divided into 10 equal parts, the blue part indicates that the test sample is one-tenth of all data sets, the 9/10 data set is called a special set whose colors are red and light blue. It is worth noting that the red part of the special data set is one-fifth of the special data set, which is used as training data. In addition, the encoding of the enhanced BSO can be seen as Fig. 3 , and encoding is divided into two parts, the first part is used as the key parameter of the support vector machine, and the remaining part is used as the attribute of the tested data.

Fig. 1.

The Flowchart of the proposed EBSO-SVM.

Fig. 2.

Data segmentation diagram.

Fig. 3.

Encoding form.

SVM is regarded as the core zone of the proposed model; in terms of the input domain, the radial basis function (RBF) kernel has been utilized to map the aggregate info into a kind of hidden layer domain. The procedure of the method deals with the main factors C and width $\gamma$, and n features subset, the first C represents the harmony among fitting error minimization and complexity of the model, the second kernel bandwidth $\gamma$ denotes the non-linear mapping action from the input domain to higher-dimensional spaces. Also, the EBSO technique considers these main parameters and the optimal feature subset at the same time. If we report more details, the continuous space has been converted using the sigmoid function into a binary domain. To this aim, the feature is selected if it be less than 0.5; otherwise, the details or gens can be deleted as exposed in Fig. 3. Finally, the optimized SVM by EBSO can give us an accurate early diagnosis of COVID-19 from the perspective of the coagulation index. In addition, the classification accuracy is set as the fitness function as Eq. 7, where average ACC refers to average test classification accuracy ratio using the SVM and the internal 5-fold CV and K is the number of folds.

$$\mathit{fitness}=\mathit{averageACC}=\frac{{\displaystyle \sum _{i=1}^k}{\mathit{testACC}}_i}{k}$$ (7)

5. Experimental designs

There are two components to the experimental phase of this analysis. The first phase will be the efficacy investigation of the proposed EBSO, and the second will use the proposed EBSO-SVM algorithm to diagnose the COVID-19 from the perspective of the coagulation index. First of all, the efficiency of the proposed EBSO is extensively verified and carried out in comparison with other algorithms on 23 benchmarks (Ballester et al., 2005), which is composed of 7 unimodal benchmark functions as shown Table 3 , 6 multimodal benchmark functions as shown in Table 4 , and 10 fixed-dimension multimodal benchmark functions as shown in Table 5 and also strictly perforce the balance and diversity analyzes of the improved EBSO and its original BSO. Several other algorithms, including BSO, DE (Price, 2013), PSO (Marini and Walczak, 2015), GSA (Rashedi et al., 2009), and MFO (Mirjalili, 2015), were involved as competitors on the common benchmark. The parameters of these peers are tested based on the original papers.

Table 3.

Unimodal benchmark functions.

Function	Dim	Range	fmin
$f_1(x)={\sum }_{i=1^n{x}_i^2}$	50	[−100, 100]	0
$f_2(x)={\sum }_{i=1^n|x_i|}+{\prod }_{i=1}^n|x_i|$	50	[−10, 10]	0
$f_3(x)={\sum }_{i=1^n{({\sum }_{j-1^i}{x}_j)}^2}$	50	[−100, 100]	0
$f_4(x)={\max }_i\{|x_i|,1⩽i⩽n\}$	50	[−100, 100]	0
$f_5(x)={\sum }_{i=1^{n-1}}[100{(x_{i+1}-x_i^2)}^2+{(x_i-1)}^2]$	50	[−30, 30]	0
$f_6(x)={\sum }_{i=1^n{([x_i+0.5])}^2}$	50	[−100, 100]	0
$f_7(x)={\sum }_{i=1^n{\mathit{ix}}_i^4}+\mathit{random}[0,1)$	50	[−1.28, 1.28]	0

Table 4.

Multimodal benchmark functions.

Function	Dim	Range	fmin
$f_8(x)={\sum }_{i=1^n-x_i\sin (\sqrt{|x_i|})}$	50	[−500,500]	−2094.9145
$f_9(x)={\sum }_{i=1^n[x_i^2-10\cos (2\pi x_i)+10]}$	50	[−5.12, 5.12]	0
$f_{10}(x)=-20\exp \left(-0.2\sqrt{\frac{1}{n}{\sum }_{i=1^n}{x}_i^2}\right)-\exp \left(\frac{1}{n}{\sum }_{i=1^n}\cos \left(2\pi x_i\right)\right)+20+e$	50	[−32,32]	0
$f_{11}(x)=\frac{1}{4000}{\sum }_{i=1^n{x}_i^2-\prod _{i=1}^n\cos \left(\frac{x_i}{\sqrt{i}}\right)}+1$	50	[−600,600]	0
$\begin{array}{c}{f}_{12}(x)=\frac{\pi }{n}\left\{10\sin \left(\pi y_1\right)\right)+{\sum }_{i=1}^{n-1}{(y_i-1)}^2\left[1+10{\sin }^2\left(\pi y_{i+1}\right)\right]+{(y_n-1)}^2+{\sum }_{i=1}^{n}u\left(x_i,10,100,4\right)\hfill \\ y_i=1+\frac{x_i+1}{4}u\left(x_i,a,k,m\right)\left\{\begin{array}{c}k\left(x_i-a\right)x_i>a\hfill \\ 0-a<x_i<a\hfill \\ k\left(x_i-a\right)x_i<-a\hfill \end{array}\right)\hfill \end{array}$	50	[−50,50]	0
$\begin{array}{c}{f}_{13}(x)=0.1\left\{{\sin }^2\left(3\pi x_1\right)\right)+{\sum }_{\mathrm{i}=1^{\mathrm{n}}{(x_{\mathrm{i}}-1])}^2}\left[1+{\sin }^2\left(3\pi x_{\mathrm{i}}+1\right)\right]+\hfill \\ \left({\left(x_{\mathrm{n}}-1\right)}^2\left[1+{\sin }^2\left(2\pi x_{\mathrm{n}}\right)\right]\right\}+{\sum }_{\mathrm{i}=1^{\mathrm{n}}u(x_{\mathrm{i}},5,100,4)}\hfill \end{array}$	50	[−50,50]	0

Table 5.

Fixed-dimension multimodal benchmark functions.

Function	Dim	Range	fmin
$f_{14}(x)={\left(\frac{1}{500}+\sum _{j=1}^{25}\frac{1}{j+\sum _{i=1}^2{\left(x_i-a_{\mathit{ij}}\right)}^6}\right)}^{-1}$	2	[−65,65]	1
$f_{15}(x)={\sum }_{i=1^{11}\left[a_i-\frac{x_1\left(b_i^2+b_i{x}_2\right)}{b_i^2+b_i{x}_3+x_4}\right]}^2$	4	[−5, 5]	0.0003
$f_{16}(x)=4x_1^2-2.1x_1^4+\frac{1}{3}{x}_1^6+x_1{x}_2-4x_2^2+4x_2^4$	2	[−5,5]	−1.0316
$f_{17}(x)={\left(x_2-\frac{5.1}{4{\pi }^2}{x}_1^2+\frac{5}{\pi }{x}_1-6\right)}^2+10\left(1-\frac{1}{8\pi }\right)\cos x_1+10$	2	[−5,5]	0.398
$f_{18}(x)=\left[1+{\left(x_1+x_2+1\right)}^2\left(19-14x_1+3x_1^2-14x_2+6x_1{x}_2+3x_2^2\right)\right]$	2	[−2,2]	3
$f_{19}(x)=-\sum _{i=1}^4{c}_i\exp \left(-\sum _{j=1}^3{a}_{\mathit{ij}}{\left(x_j-p_{\mathit{ij}}\right)}^2\right)$	3	[1,3]	−3.86
$f_{20}(x)=-\sum _{i=1}^4{c}_i\exp \left(-\sum _{j=1}^6{a}_{\mathit{ij}}{\left(x_j-p_{\mathit{ij}}\right)}^2\right)$	6	[0,1]	−3.32
$f_{21}(x)=-\sum _{i=1}^5{\left[\left(X-a_i\right){\left(X-a_i\right)}^{\mathrm{T}}+c_i\right]}^{-1}$	4	[0,10]	−10.1532
$f_{22}(x)=-\sum _{i=1}^7{\left[\left(X-a_i\right){\left(X-a_i\right)}^{\mathrm{T}}+c_i\right]}^{-1}$	4	[0,10]	−10.4028
$f_{23}(x)=-\sum _{i=1}^{10}{\left[\left(X-a_i\right){\left(X-a_i\right)}^{\mathrm{T}}+c_i\right]}^{-1}$	4	[0,10]	−10.5363

The experimental results EBSO-SVM, employed for a diagnosis of coagulation indices in COVID-19 in the field of data collection, was extensively reviewed by the proposed EBSO method for optimizing the combination of the best parameter and SVM sub-subset function. Several typical learning procedures, including original SVM, GWO-SVM, MFO-SVM, PSO-SVM, ELM, and KNN, were also compared in the case of EBSO-SVM. The two main parameters of $[-2^{\wedge 6},2^{\wedge 6}]$ and $[-2^{\wedge 6},2^{\wedge 6}]$, respectively, have been specified in the original SVM. In order to avoid the uncertainty in tests due to the large datasets, all info has been scaled into the interval of [-1, 1].

The MATLAB tests has been performed with the Xeon CPU E5-2660 v3 (2.60 GHz) and 16 GB of RAM on a Windows Server 2018 R2 operating machine. A Cross-Validation 10-fold (CV) is used to assess classifying findings in an unbiased and objective manner. In addition, the EBSO-SVM performance assessment was carried out using four normal parameters comprising specificity, sensitivity, classification accuracy (ACC), and a Matthews (MCC). The detailed definition of the formula can refer to Wang and Chen (2020).

6. Experimental and analytical results

6.1. Results of benchmark functions

In this part, the 23 benchmarks have been utilized to evaluate EBSO’s efficiency, and these cases have always been applied in multiple projects using fair comparison rules (Wang et al., 2020, Weng et al., 2021, Qiang et al., 2020, Yang et al., 2018, Zhao et al., 2020, Zou et al., 2019). Furthermore, 30 experiments were performed independently to limit the impact of random variables. The EBSO is compared to the BSO, DE, PSO, GSA und MFO algorithms in this investigation. Thirty individual executions carried out all of these ways to these benchmark standards.

In the Table 6 , detailed experimental results are demonstrated by the average and standard deviationSTDvalues. It can be seen that the EBSO have the best property of not only average values but also STD, though it may perform not well on some cases Moreover, the EBSO is substantially higher than the other couples in the most relevant test criteria in the Wilcoxon test. Table 7 presents Friedman$\prime s$ EBSO test results against all other rivals. The EBSO shows the first best results of these benchmarks, with the poorest findings being BSO, MFO, GSA, PSO, and DE, in line with the average ranking of the algorithms in question. One cause for this may be the major aspects of the existing movement strategies abstracted from optimizing harris hawks. In this analysis, the original BSO can be enhanced between exploration and mining.

Table 6.

The statistical experiment results and the comparison algorithms on the test benchmarks.

Algorithms	F1		F2		F3		F4		F5
	mean	STD	mean	STD	mean	STD	mean	STD	mean	STD
EBSO	4.0409869	4.565512	0.848937	0.32468	5308.9514	1916.3288	7.577523	2.163198	909.548	672.3266
BSO	14216.69366	1936.4153	73.042895	5.8637549	70294.044	10991.059	70.005317	4.7751876	14888839	5430531.7
DE	286.5318805	80.04938	6.7574347	0.6717474	85066.864	8903.3004	59.555309	3.056055	72392.185	18945.928
PSO	331.5727461	37.50266	1661020.4	4334397.6	4836.1655	1315.6191	10.840428	1.829634	681313.26	166072.15
GSA	40.21357786	12.449257	30.790421	7.9833557	3082.485	1262.17	18.862995	2.798129	53327.901	28552.84
MFO	14353.41746	8511.954	63.966444	19.574595	62399.277	13471.362	82.965428	4.3295102	24824531	25784078
Algorithms	F6		F7		F8		F9		F10
	mean	STD	mean	STD	mean	STD	mean	STD	mean	STD
EBSO	2.4478589	0.942465	0.229856	0.10233	−11173.4	718.747	172.6646	17.44842	1.989142	0.515402
BSO	14572.66347	2369.2804	9.0611144	2.8662241	−8879.693	666.81149	266.02169	15.214212	16.952533	0.508035
DE	273.4697128	64.452586	0.4612938	0.1143467	−8907.786	363.95443	231.08217	11.055591	5.5215554	0.4519889
PSO	339.425991	36.361857	254.76347	37.413359	−7155.622	1294.4444	566.25084	34.593948	10.276171	0.3276113
GSA	32.75943606	5.2913592	110.65906	16.96339	−3077.189	664.71759	335.0435	13.968292	4.9673752	0.3732885
MFO	11176.15861	4626.9133	15.377483	14.019924	−10395.76	1185.279	289.99885	44.965597	19.565532	0.3733107
Algorithms	F11		F12		F13		F14		F15
	mean	STD	mean	STD	mean	STD	mean	STD	mean	STD
EBSO	0.9543863	0.121121	7.0724065	3.7750497	26.51097	11.10599	0.998004	1.96E-16	0.0043187	0.0084562
BSO	133.7178699	14.668046	13245314	6397676.3	41034266	19312164	0.9980038	1.877E-11	0.0015329	0.0009859
DE	3.975427709	1.0276143	28.112923	13.85836	6299.2424	9589.2105	1.5904907	1.5560234	0.001011	0.000272
PSO	1.15537423	0.0262132	12.926533	2.8005088	6343.4568	8514.9991	4.2516365	2.5355863	0.0016576	0.0007102
GSA	2.694441032	1.1526546	6.366073	2.89656	2004.7167	3433.8448	1.6061734	0.6890466	0.0052248	0.0027588
MFO	105.8181227	61.608213	19925243	13625159	125560603	172385521	6.090624	4.2799447	0.0024782	0.0031255
Algorithms	F16		F17		F18		F19		F20
	mean	STD	mean	STD	mean	STD	mean	STD	mean	STD
EBSO	−1.03162845	9.735E-16	0.397887	0	3	1.39E-15	−3.86278	6.94E-16	-3.286327	0.0574308
BSO	−1.0316284	3.13E-07	0.3978874	5.085E-10	3	2.94E-09	−3.86278	1.56E-08	-3.301074	0.0356608
DE	−1.03162845	1.958E-16	0.3978874	0	3	1.92E-15	−3.86278	7.83E-16	−3.32183	0.000367
PSO	−1.02490226	0.0053364	0.4117694	0.0142042	3.6032981	0.8927914	-3.808387	0.0557767	-2.643092	0.1640719
GSA	−0.99227955	0.0216461	0.408267	0.0101211	3.3711742	0.3749822	-3.838708	0.0146349	-2.802657	0.2136215
MFO	−1.03162845	1.282E-16	0.3978874	0	3	2.5E-15	−3.86278	8.24E-16	-3.237214	0.0770919
Algorithms	F21		F22		F23
	mean	STD	mean	STD	mean	STD
EBSO	−8.6538929	3.16084	−8.443763	3.2038506	−10.5364	3.14E-12
BSO	−7.99693838	2.7755981	−8.72163	1.61663	−8.203202	2.4544712
DE	−8.3009019	2.0025497	−10.16796	0.5866285	−9.74481	2.1880623
PSO	−2.58648777	1.2965977	−2.450181	1.0838472	−3.096436	1.4694115
GSA	−2.61413202	0.5856597	−2.532658	1.0300769	−2.80029	0.7462128
MFO	−5.39182791	3.4275629	−7.147333	3.520498	−7.47797	3.9669906

Table 7.

The results of Friedman’s test.

	EBSO	BSO	DE	PSO	GSA	MFO
mean level	1.4236	5.4251	3.3625	3.5489	4.3549	4.5219

Fig. 4 shows the convergence curves of these algorithms for several selected benchmarks to verify the performance of the designed EBSO. From this viewpoint, the planned EBSO indicates the quick capacity of the conference and the apparent superiority to all other competitors in this benchmark. Furthermore, the developed EBSO contains rapid convergence searches, such F7, F21, and F23, which ensure it achieves an optimal theoretical value in no time. In addition, the same convergence tendency is found as regards other benchmarks. In short, it can be concluded that the property of the original BSO can be significantly enhanced.

Fig. 4.

Convergence curves of selected benchmark functions.

6.2. Application in the diagnosis of COVID-19 from the perspective of coagulation index

In this part, the proposed algorithm EBSO-SVM for diagnosis of COVID-19 from the perspective of coagulation index is evaluated deeply. Table 8 records the detailed statistical results of the collected COVID-19 coagulation index data set. The 91.9195% classification accuracy of the ESMA-SVM can be seen from this Table 8, 90.529% of Matthew correlation coefficient, 90.9912% of sensitivity, 88.5705% of specificity, and their variance is 0.049967, 0.039574, 0.050529 and 0.060838 respectively. In addition, the suggested EBSO-SVM can acquire the optimum SVM model configuration automatically, largely due to the upgraded EBSO, which can effectively detect optimum parameters settings and feature subset.

Table 8.

The results of EBSO-SVM on collected data.

Fold	ACC	MCC	Sensitivity	Specificity
#1	0.88759	0.85847	0.95214	0.80251
#2	0.88952	0.89568	0.89687	0.87548
#3	0.97521	0.96324	0.96252	0.87549
#4	0.97635	0.84356	0.88573	0.85321
#5	0.98521	0.88821	0.96595	0.97584
#6	0.88521	0.95321	0.89654	0.93251
#7	0.88597	0.89524	0.90251	0.96524
#8	0.96874	0.88754	0.80521	0.81258
#9	0.87584	0.93254	0.95624	0.84265
#10	0.86231	0.93521	0.87541	0.92154
Mean	0.919195	0.90529	0.909912	0.885705
STD	0.049967	0.039574	0.050529	0.060838

Furthermore, the EBSO-SVM approach is comparable with the original SVM and other evolutionary SVM computational methods (including BSO-SVM, Original SVM, GWO-SVM, MFO-SVM, PSO-SVM, and two popular ELM and KNN algorithms). A complete statistical experiment in Table 9 shows comparisons of the accuracy, Matthew’s correlation coefficients, susceptibility, specificity, and standard deviation. The result can be overlooked that in quatre assessment methodologies such as ACC, MCC, sensibility, and specificity, the EBSO-SVM algorithm is higher than other competition and its related standard difference between all models is also less relevant. The EBSO-SVM is more efficient and stable compared to the traditional CBSO-based SVM. It should be noted that the original SVM, original ELM, and KNN all demonstrate their worst diagnosis performance of COVID-19 from the coagulation index perspective, which can preliminarily be shown to improve the SVM model selection ability and the ability to solve the precise diagnosis of COVID-19 from the capacity perspective from the algorithm suggested in this article. We can observe in this experiment that EBSO SVM can automatically obtain the best property among all these competing designs, mainly thanks to the upgraded EBSO, which automatically identifies the best SVM parameters and the perfect sub-features.

Table 9.

The statistical experiment results of comparison in terms of the four metrics.

Algorithms	ACC	MCC	Sensitivity	Specificity
EBSO-SVM	0.91919 ± 0.04996	0.90529 ± 0.03957	0.90991 ± 0.05052	0.88570 ± 0.06083
BSO-SVM	0.88574 ± 0.07541	0.86354 ± 0.06584	0.88579 ± 0.07584	0.83595 ± 0.09658
SVM	0.79254 ± 0.08025	0.77854 ± 0.08547	0.79587 ± 0.09521	0.70124 ± 0.16254
GWO-SVM	0.85362 ± 0.06524	0.83569 ± 0.06854	0.87623 ± 0.07215	0.86598 ± 0.14584
MFO-SVM	0.84325 ± 0.06758	0.83758 ± 0.06851	0.88215 ± 0.07219	0.85784 ± 0.09236
PSO-SVM	0.81587 ± 0.06741	0.85812 ± 0.06954	0.87259 ± 0.07359	0.86325 ± 0.10594
ELM	0.80563 ± 0.08652	0.76325 ± 0.08812	0.78851 ± 0.10584	0.71265 ± 0.13625
KNN	0.78954 ± 0.08547	0.80145 ± 0.10258	0.78852 ± 0.11263	0.73541 ± 0.14587

In addition, the developed EBSO will be used to simultaneously optimize parameters and choose functions for SVM to diagnose COVID-19 from the coagulation index perspective. The 10-fold CV approach during function selection is used in this work. Table 10 shows the detailed number of features, and statistical values picked for each 10-fold cycle. Regarding statistics, AGE, INR, PTA, and D-D features were selected respectively by the EBSO-SVM with values 9, 8, 9, and 9 while comparing the other features with the features with fewer ones. The EBSO-SVM is superior to other competitors. But other opponents did not fulfill these characteristics. It can thus be concluded that those characteristics that typically seem present can recognize COVID-19 early and discriminate against other low-frequency characteristics. In practice, more consideration for these features AGE, INR, PTA, and D-D should therefore be given in relation to the underlying information in these frequency characteristics.

Table 10.

The numbers of selected feature.

Index	Algorithms
	EBSO-SVM	BSO-SVM	GWO-ELM	MFO-ELM	PSO-ELM
F1	0	0	0	2	3
F2	9	6	7	7	8
F3	2	2	1	4	3
F4	8	8	7	8	7
F5	9	8	7	7	8
F6	1	3	6	2	4
F7	0	2	2	3	1
F8	1	2	1	1	2
F9	9	8	8	7	9

In addition, the comparison results among these methods regarding CPU time consuming via 10-fold CV is shown in Fig. 5 . It can be observed that the original ELM has the least CPU time whose speed is the fastest among these algorithms, the original SVM has the second least time. An explicit argument is that a lot of execution time will be saved without the aid of search algorithms compared to models. However, the unintended outcome is a considerably decreased algorithm classification performance. It can also be observed that the presented EBSO-SVM performs only the fourth-least time consuming, which has more time than the original BSO-SVM, and it illustrates that the add operator increases the algorithm’s execution time. The time-consuming of KNN and GWO-SVM is very similar and, the PSO-SVM has the most time-consuming. Finally, although the proposed EBSO-SVM has good classification performance, its performance increase is at the cost of time consumption, and this also directs us towards the future path of study using acceptable parallel programming to reduce the duration for ECPA-KELM CPU use.

Fig. 5.

Comparison results among these methods in terms of CPU time via 10-fold CV.

7. Discussion

Early recognition and discrimination of COVID-19 severity from a coagulation index standpoint were studied using EBSO to optimize the parameter and select SVM features simultaneously. Important aspects such as Age, INR, PTA, and D-D have been found. Through the coagulation index, an EBSO-SVM model was subsequently designed to diagnose COVID-19 accurately. We believe that the EBSO-SVM model could contribute to informing the clinical decision-making process.

Several studies have shown that age plays a key role in developing Severe Acute Respiratory Syndrome (SARS) and Middle East Respiratory Syndrome (MERS). Meanwhile, researchers confirm that advanced age was associated as a strong independent risk predictor of SARS and MERS (Majumder et al., 2015, Choi et al., 2003). Similarly, investigators have demonstrated that age plays a critical role in the progression of the COVID-19 disease in this outbreak. Chen and colleagues revealed that age was positive association with the degree of radiographic severity in patients with COVID-19 (Chen et al., 2020). Many researchers have found that older patients are more vulnerable to COVID-19 infections and that they are more prone to develop severe COVID-19 complications (Wang et al., 2020). Wang et al. reported that adults are more vulnerable and susceptible to COVID-19 than children and the adult’s condition was severe than that in children (Wang et al., 2020). Wang et al. also found that these elderly persons, once infected with SARS-CoV-2, are prone to death (Wang et al., 2020). One of the possible reasons is that with aging, the cellular and humoral immune function of the body gradually declines, resulting in an increased risk of infection (Weiskopf et al., 2009, Opal et al., 2005). Another possible reason is that elderly patients with numerous comorbidities, including hypertension, diabetes, chronic lung diseases, may be more prone to acute respiratory failure and have a poor prognosis (Liu et al., 2020, Zhou and Liu, 2020). Consistent with previous studies, we also found the average age of the severe COVID-19 group was 61.43 ± 17.64 years, and that of the non-severe COVID-19 group was 42.30 ± 11.53 years, suggesting age is an important factor affecting rehabilitation. To summarize, age may be a promising indicator for predicting the prognosis of patients with COVID-19.

Severe infection and inflammation lead to activation of coagulation. Research findings reveal that coagulation function abnormalities are commonly found in patients with sepsis. Fibrinogen levels of C-reactive protein were closely correlated to the sepsis, suggesting the level of fibrinogen can reflect the inflammatory state of the body(Kim et al., 2007). D-dimer is a fibrin degradation product of the crosslinked fibrin polymer under the fibrinolytic system (Bai et al., 2017). D-dimer is the smallest fragment of fibrin degradation products and serves as an important molecular marker for fibrinolytic system hyperactivity and hypercoagulable state in vivo (Baboolall et al., 2019). The level of D-dimer in normal human serum is less than 280 ng/mL, and the concentration is difficult to detect in normal situations (Li et al., 2018). D-dimer plays an important role in inflammation. When the body is infected, a variety of inflammatory factors may damage vascular endothelial cells, leading to activation of blood coagulation and consume mass clotting factors inducing hyper coagulability (Duarte et al., 2015). Guneysel et al. found that plasma D-dimer levels are positively correlated with the degree of community-acquired pneumonia (CAP). Moreover, patients with severe CAP had significantly higher plasma D-dimer levels than those with a non-severe CAP suggesting D-dimer may be a promising prognostic indicator in CAP (Guneysel et al., 2004). It has been reported that various vasoactive substances were increased in patients with CAP or mechanical ventilation occurs more easily when plasma D-dimer levels increase (Li et al., 2018). Recently, a population-based study of more than 300 CAP patients showed that plasma D-dimer levels are measured before the use of antibiotics, suggesting plasma D-dimer levels are positively associated with the acute physiology and chronic health evaluation II (APACHE II) and pneumonia severity index (PSI) scoring system. Moreover, the D-dimer levels in the non-survival CAP patients were significantly higher than in the survival CAP group (Querol-Ribelles et al., 2004). In line with these findings, Han et al. found that the levels of plasma D-dimer in patients with severe COVID-19 were significantly higher than those in the non-severe COVID-19 group, and monitoring plasma D-dimer levels are helpful for the early diagnosis, management of severe/critically ill patients (Han et al., 2020). In addition, among COVID-19 patients, high levels of D-dimer were closely related to the poor prognosis of patients with COVID-19 (Tang et al., 2020). Similar to their findings, in this study, we revealed that plasma D-dimer levels were 4.14-fold higher in the severe COVID-19 group compared with the non-severe COVID-19 group (P = 0.032), suggesting that plasma D-dimer levels may be regarded as a promising indicator of clinical outcome in COVID-19 patients.

Similar to D-dimer, PT is a sensitive screening test for the extrinsic coagulation pathway and a vital monitoring indicator in clinical anticoagulation therapy (Liu et al., 2018). Meanwhile, PT is also an important prognostic index for patients with severe hepatitis. PTA has the same significance as PT. Coagulation function indicators are believed to be correlated with the clinical course or prognosis of sepsis. PT, APTT, FIB, and D-dimer are widely used for detecting sepsis (Hoshino et al., 2017). A previous study has revealed that PT was prolonged in patients with sepsis, at the same time, proving PT was positively related to illness severity and poor prognosis in sepsis patients (Walsh et al., 2010). Vasques and colleagues reported that coagulation disorder has a great influence on sepsis, especially septic shock. The prognosis became worse with the increase of PT (Macrae et al., 2014). Therefore, PT can be served as a valuable prognostic indicator in sepsis and septic shock. Additionally, a population-based study of 113 deceased COVID-19 patients also suggested that all fatal cases had sepsis complications, suggesting sepsis is the most common and serious complication of exacerbation of covid-19 (Chen et al., 2020). In terms of PT, PTA, Liu et al. observed a very significant difference between severe COVID-19 and non-severe COVID-19 patients (Liu et al., 2020). In our study, we found that the average PT of the severe COVID-19 group was 14.93 ± 3.05 s, and that of the non-severe COVID-19 group was 13.03 ± 0.78 s, and the average PTA was 88.28 ± 8.40 % in the severe COVID-19 group and 95.88 ± 7.35 % in the non-severe COVID-19 group. This finding was similar to the results from previous studies (Wang et al., 2020, Liu et al., 2020). In short, a close and important relationship exists between COVID-19, sepsis, and coagulation disorder. There are very few reports in the literature describing coagulation function indexes and clinical parameters to predict the severity and prognosis of the infectious disease jointly. As far as we know, this is the first attempt at combining age, PT, PTA, and D-dimer for predicting COVID-19 severity using the machine learning method.

Based on the finding of this research on the performance of the proposed hybrid support vector machine, we can propose its application to more variety of complex problems, such as active surveillance (Pei et al., 2020), evaluation of human lower limb motions (Qiu et al., 2016), prediction (Xu et al., 2020), medical diagnosis (Chen et al., 2020, Fei et al., 2020, Hu et al., 2021, Saber et al., 2021). Also, it can be utilized to evaluate the service ecosystem (Xue et al., 2020, Xue et al., 2019), optimal e-healthcare systems (Zhang et al., 2020), optimal performance design (Meng et al., 2018), edge computing (Sheng et al., 2021, Zhiang et al., 2020), drug discovery (Zhuo et al., 2020), optimal control (Luo et al., 2020, Qian et al., 2021, Ye et al., 2020, Zhao et al., 2021), and image and video processing (Mai et al., 2018, Zhou et al., 2021). Another potential is to develop the binary variant of the optimizer for dealing with wrapper-based feature selection (Fan and Zhang, 2021, Zhang et al., 2020, Zhang et al., 2015).

However, the present study has some limitations. Currently, our data set came from a retrospective design and single-center data, and the sample size is not big enough. Larger study samples and prospective studies are needed in order to further improve and optimize our model. Second, independent/ external datasets will still be required to train the EBSO-SVM model in the future so as to make the model more reliable. Third, we recommend including more indicators of hematological parameters such as blood routine, liver function indicators, kidney function indicators, immune index, inflammatory indicators, and blood gas indexes in future studies. In addition, the proposed EBSO-SVM may still have some room to improve. The performance of the algorithm proposed is obtained at the expense of some computational costs. For this, in the future, we will consider paralleling (Zhiang et al., 2017) the implementation of the algorithm to reduce the consumption of computational time. In addition, the algorithm proposed EBSO-SVM should seek more medical application scenarios and try to solve more medical classification problems.

8. Conclusion and future Work

The work uses clinical data of the affiliated Yueqing Hospital of Wenzhou Medical University to develop an effective EBSO-SVM approach to identify and discriminate COVID-19 severity early on (Yueqing, China). The fundamental novelty in the proposed methodology is the incorporation of the new strategy by the present EBSO to improve and restore search capabilities to the original BSO. The efficiency of EBSO was thoroughly controlled with the 23 benchmark tests compared to many other competitors. The experimental findings showed that the proposed EBSO is significantly better able than other partners to achieve this feature optimization. In addition, the recommended EBSO was utilized to develop synchronized optimal SVM parameters and features; the resulting EBSO-SVM was successfully applied to early identification and discern its severity with the COVID-19. An EBSO-SVM analysis with other competitive algorithms has also been carried out. Moreover, the findings demonstrated EBSO-SVM to forecast the more stable attribute more accurately.

A number of concerns must be examined in more detail for future study. In order to lessen the computing burden in the application phase, more variables and coefficients are introduced, and the following parallel processing should be highlighted. Readers can also collect additional data samples to construct a safer and more effective system. In order to extend the application of this technique, EBSO-SVM can also be utilized for prejudging other conditions like the clustering and splitting of the picture into CT.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.