key: cord-0737839-l0fc2gef
authors: Shi, Beibei; Ye, Hua; Zheng, Long; Lyu, Juncheng; Chen, Cheng; Heidari, Ali Asghar; Hu, Zhongyi; Chen, Huiling; Wu, Peiliang
title: Evolutionary warning system for COVID-19 severity: Colony predation algorithm enhanced extreme learning machine
date: 2021-07-30
journal: Comput Biol Med
DOI: 10.1016/j.compbiomed.2021.104698
sha: c34cb333b4c5cf2bad473539ab8bd2999b71f896
doc_id: 737839
cord_uid: l0fc2gef

Coronavirus Disease 2019 (COVID-19) was distributed globally at the end of December 2019 due to severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). Early diagnosis and successful COVID-19 assessment are missing, clinical care is ineffective, and deaths are high. In this study, we investigate whether the level of biochemical indicators helps to discriminate and classify the severity of the COVID-19 using the machine learning method. This research creates an efficient intelligence method for the diagnosis of COVID-19 from the perspective of biochemical indexes. The framework is proposed by integrating an enhanced new stochastic called the colony predation algorithm (CPA) with a kernel extreme learning machine (KELM), abbreviated as ECPA-KELM. The core feature of the approach is the ECPA algorithm which incorporates the two main operators that have been abstained from the grey wolf optimizer and moth-flame optimizer to improve and restore the CPA research functions and are simultaneously used to optimize the parameters and to select features for KELM. The ECPA output is checked thoroughly using IEEE CEC2017 benchmark to verify the capacity of the proposed methodology. Finally, in the diagnosis of COVID-19 using biochemical indexes, the designed ECPA-KELM model and other competing KELM models based on other optimization are used. Checking statistical results will display improved predictive properties for all metrics and higher stability. ECPA-KELM can also be used to discriminate and classify the severity of the COVID-19 as a possible computer-aided method and provide effective early warning for the treatment and diagnosis of COVID-19.

Coronavirus Disease 2019 (COVID-19) was found to have spread around the world in late December 2019 as a result of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) [1, 2] . COVID-19 has shown an intensive global spread, and thus the danger to human health is serious. As of 11 June 2020, COVID-19 was responsible for 7,273,958 confirmatory cases worldwide and 413,372 deaths [3] . COVID-19's clinical features may echo other coronavirus diseases, including Severe Acute Respiratory Syndrome Coronavirus (SARS-CoV). Globally, the current human mortality rate for infection with SARS-CoV-2 is 3.4% [4] . However, the mortality rate of serious COVID-19 patients in Wuhan was as high as 10% − 40% [5] . Therefore, early and accurate identification of severe COVID-19 pa-medical field. Similarly, machine learning-based technology has been applied to in disease diagnosis [18] . Computer tomography (CT) or x-ray image recognition [19, 20] , disease epidemic, surveillance, and control [21, 22] in the course of the COVID-19 outbreak. Ebadi et al. [23] used multiple sources (PubMed and ArXiv) to define the scene of the current COVID-19 research with multiple learning machines by identifying latent subjects and analyzing the evolution, the similarities between publications, and sentiment of the research topics developed. Three NLP algorithms have been developed to trace positive CT imaging of typical SARS-CoV-2 viral inflammatory diseases [24] . A recent data collection of COVID-19 CT scans, known as COVID-CT-MD, consists not only of COVID-19 cases but also stable and community-acquired pneumonia (CAP) participants that can assist in developing advanced machine learning and DNN solutions [25] . Chowdhury et al. [26] developed an early warning method to predict mortality risk in machine learning COVID-19 patients. The random vulnerability model used by the forest machine for the creation of a novel COVID-19 Vulnerability Index (C19VI) [27] . In brief, machine learning can be extremely efficient in the analysis of COVID-19 before diagnosis and disease data.

This research is the first time an evolutionary kernel extreme learning machine is built to diagnose COVID-19 from a biochemical index perspective.

For the first time, a system to diagnose COVID-19 from a biochemical index point of view is designed for an evolutionary kernel extreme learning machine in this study. The two core operators use this proposed method (ECPA-KELM) to improve and re-establish the search capabilities for colony predation algorithm (CPA), abstracted from the Grey Wolf and Moth Flame Optimizers, which can provide for considerable convergence and the potential to spring from the stagnant local population, ECPA is designed to simultaneously perform diagnosis of COVID-19 parameter optimization and selection features of KELM. The ECPA output is first thoroughly verified through IEEE CEC2017 benchmark test cases [28] to verify the capacity of the proposed methodology. Lastly, COVID-19 clinical data was used with biochemical indexes to develop the ECPA-KELM and other competitor KELM models based on other optimization algorithms. By analyzing the experimental findings, the ECPA core compensation is verified, and a solid ECPA-KELM in terms of various performance assessment indexes to determine the COVID-19 status can be achieved. The results of the test showed that the ECPA-KELM proposed was seemingly beneficial.

The key contribution in this analysis is as follows:

• To improve and restore the CPA search capability (ECPA), both core operators have taken away the gray wolf optimizer and moth-flame optimizer.

• The proposed hybrid ECPA has achieved a significant impact on CEC2017 optimization tasks.

• For the first time, the ECPA proposed successfully solved KELM's parameter optimization and feature selection simultaneously.

• An effective ECPA-KELM technique is used to help diagnose COVID-19 from the perspective of biochemical indicators.

The paper was organized accordingly-the materials and procedures described in paragraph 2. The proposed ECPA algorithm is presented in Section 3. Section 4 describes the proposed ECPA-KELM model. Section 5 describes the designs of the experiments. Section 6 displays the results of CEC2017 ECPA and the diagnosis of COVID-19 data set simulations and ECPA-KELM. Section 7 deals with the results. The conclusion and trajectory of the future are shown in Section 8.

The study included COVID-19 patients at Wenzhou Medical University affiliated Yueqing Hospital, Wenzhou, China. The research was accepted by the Ethics Committee of affiliated Yueqing Hospital, Wenzhou Medical University (No. 202000002 Ethics), and by all COVID-19 patients, an informed consent document was signed. A total of 51 COVID-19 patients were included in the analysis in retrospect between January 21 and March 20, 2020. For each COVID-19 patient, information on gender, age, biochemical index, and blood electrolyte was collected. Biochemical indexes and blood electrolytes were determined using an automated biochemical analyzer (BS-190; Mindray, Shenzhen, China) in the laboratory of clinical biochemistry, the Affiliated Yueqing Hospital of Wenzhou Medical University.

In our research, COVID-19 was diagnosed based on criteria developed by the Peoples' Republic of China National Health Commission. Once the diagnosis of COVID-19 was confirmed, we divided patients into four categories according to the clinical manifestations: mild, general, severe, and critically ill patients. Clinical characteristics of mild COVID-19 patients include no symptoms or mild symptoms, no lung involvement. Clinical characteristics of general COVID-19 patients include respiratory symptoms (fever, dry cough, fatigue, nose congestion, runny nose, sore throat), gastrointestinal symptoms (nausea, vomiting, diarrhea), and pulmonary disease SARS-CoV-2. Clinical criteria for severe COVID-19 patients include at least one of the following: a) patient exhibit dyspnea and respiratory rate ≥ 30 breaths/minute; b) the levels of blood oxygen saturation ≤ 93%; c) the oxygenation index (OI) ≤ 300 mmHg. Clinical criteria for critically ill COVID-19 patients include at least one of the following: a) patient develop acute respiratory failure requiring mechanical ventilation; b) patient develop shock; c) patient present with multiple organ failure requiring treatment in an intensive care unit (ICU). Mild, general COVID-19 patients were categorized into one group and named nonsevere COVID-19 group, and severe, critically ill patients were categorized into one group and named severe COVID-19 group. 

Statistical analysis using SPSS software was performed. Age, biochemical index, and blood electrolyte comparison were tested using an independent-samples t-test between the non-severe COVID-19 and severe COVID-19 groups. Age, biochemical index, and blood electrolyte are presented as mean ± standard deviation (̄ ± ) for continuous variables. The patient information, biochemical index, and blood electrolyte of COVID-19 patients have listed in Table 1 . The statistical results of age, biochemical index, and blood electrolyte were shown in Table 2 .

Optimization methods, in addition to other cases in healthcare systems, have found their value and obtained great attention in many fields such as scheduling problems [29, 30] , image segmentation [31, 32] , fault diagnosis of rolling bearings [33, 34] , bankruptcy prediction [35, 36, 37] , wind speed prediction [38] , engineering design problems [39, 40, 41, 42] . Also, they have shown more variety of potentials in the hard maximum satisfiability problem [43, 44] , param-eter optimization [45, 46, 47, 48] , PID optimization control [49, 50, 51] , gate resource allocation [52, 53] , feature selection [54, 55, 56, 57, 58] , medical data classification [59, 60, 61, 62] , detection of foreign fiber in cotton [63, 64] , and prediction problems in educational field [65, 66, 67, 68, 69] . One of the recent methods is CPA. The dominant idea of CPA is the predation process and predation strategy of group of hunting animals. The algorithm mimics the supportive behavior of social animals and the characteristics of selective hunting. Thus this algorithm is based on the coexistence of social animals. The main steps are composed of communication and collaboration, disperse food, encircle food, supporting closest individual, and searching for the food. The following formulas represent individual cooperative communication and food searching behavior:

where r is in the range of [0,1], ⃗ 1 ( ) is the individual looking for food, ⃗ 1 and ⃗ 2 are the two closest to prey in the j-th dimension, ∈ 1, 2, … , , ⃗ ( + 1) is the latest updated position of the individual. Here, the predation strategy displayed by individuals in search is simulated mathematically as ⃗ ( + 1) = ⃗ − ⋅ 1 ( − ) + , where ⃗ ( + 1) is the position of a population and ⃗ is the position of food, S represents the strength of prey, and its absolute value decreases from a to 0 with the number of evaluations, 1 is the 1 ; 2 ; 3 ; … , = represents the dimension of the population, ub, and lb represent the upper and lower bounds. It should be noted that S is as follows:

where N is the number of individuals, 0 decreases from a to 0 with the number of evaluations and t represents the current number of evaluations, 2 is a random number of [0,1]. The hunting group will surround the single prey and keep approaching the prey. This stage can be represented by mathematical simulation as follows:

where D is the distance between the current individual and the prey as

. Mathematical formulaes describe the probabilities of implementing these two predatory strategies as below:

Since the group can experience problems in hunting beasts, the closest person calls for peer help. In mathematical formulas, its policy can be expressed as ⃗ ( + 1) = ⃗ , where ⃗ nearest is the location of the nearest predator in the support group, ⃗ is the predator near the prey nearby. The searching for the food can be as:

where 1 denotes the distance of random group movement, 4 is a random number of [0,1], and ⃗ is a new individual position formed randomly by individuals. See original paper for detailed details [70] .

One core operator is a grey wolf optimizer consisting of social hierarchy, surrounding beasts, hunting, attacking prey and searching for projection pieces. The key concept is the use of hierarchy. This is the core mathematical model∶

where the ⃗ means the distance from the prey to the grey wolves, ⃗ and ⃗ are vectors of the coefficient ⃗ = 2 ⃗ ⋅ ⃗ 1 − ⃗ and ⃗ = 2 ⋅ ⃗ 2 , ⃗ , ⃗ and ⃗ are three wolves nearest to the current prey, the represents the present version. You may refer to more info [71] .

⃗ ( + 1) =

Another core operator is a moth-flame Another main operator is the moth-flames optimization (MFO) [72] . The location of each moth is changed to a flame with the following equation in order to mathematically model the conduct of moths: = ( ), where is the ith moth, means the jth flame, and S is the spiral function. For the MFO algorithm, a logarithmic spiral is described below:

where specifies the ith moth distance for the jth flame, b is a logarithmic spiral constant for determining the form, and t is a [ 1, 1] random number, D is obtained as = | − |, where i indicates the ith moth, j indicates the jth flame, and i is the distance of the ith moth for the jth flame. More detailed information can be seen in [72] .

Extreme learning machine (ELM) [73] can learn fast and has very few adjustments parameters, and does not provide the option of input weights and secret preconditions as the latest learning algae to feedforward neural networks in a First Author et al.: Preprint submitted to Elsevier single hidden layer. The new theory for an extreme learning machine from the kernel has recently been extended (KELM) [74] . The following is a short overview of the KELM method:

Training set as = {( , )| ∈ , ∈ , = 1, 2, ..., } is given, where n × 1 input feature vector of and m× 1 of target vector. an activation function ℎ( ) can be modeled as follows:

The is the weight vector between the hidden layer of the ith and the input layer, and the distinctiveness between the hidden layer of the ith is called , the weight vector between the ith and the output layer is ; is the target vector of the jth input data.

⋅ is the outcome of the and internal product. Thẽ means the number of the hidden layer nodes. To assess these samples correctly,

, , , which can be given as = , where

H is the result matrix of the hidden layer neural network, with the ith column of H being the ith hidden output neuron with respect to 1 , 2 , … , . Enter weights and vector secret layer bias need not be adjusted. On this basis, the output weight of the linear system = can be given mathematically with the least square solution̂ :

According to the Moore-Penrose (MP) generalized inverse and the kernel learning theory, the output function of KELM is shown as follows:

Usually radial basis function kernel are used in this models as ( , ) = (− || − || 2 ), the penalty parameter C and gamma kernel µ are the two main parameters, based on the definitions above. The first parameter C can better off the minimization of fitting errors and the complexity of the model. The second parameter, , determines the nonlinear projection from the input area to a large space. More detailed information on KELM can be shown in [74] .

As we know, in many decision-making fields, algorithms are required to provide an approximate optimal solution [75, 76, 77] . For such complex calculation problems, evolutionary computing methods are widely used to solve such problems. Such as, The global algorithm of particle swarm is used to solve gravity abnormality parameters [78] , hybrid genetic algorithm for diagnosing heart disease, and fluid logic classification [79] . Among these evolutionary computing algorithms, there are more widely used particle swarm optimization [80] , genetic algorithm [81] . Furthermore, recently, several excellent evolutionary calculation methods have been proposed, grasshopper optimization algorithm [82] , harris hawks optimization [83] , hunger games search [84] . Among all these evolutionary computing algorithms, CPA, as of the most recent participant and Very easy to implement and complete the application, has been proposed by our team [70] . Through strict experimental data statistics and the application of several engineering problems, the algorithm's efficiency has been tested strongly.

In this analysis, ECPA is designed to increase and restore the search capabilities of the original CPA, both core operators abstracted by the gray wolf optimizer and mothflame optimizer. To the best of our knowledge, the CPA is integrated with multiple efficient operators for the first time. The designed algorithm ECPA can be divided into three parts. The first is to perform the initialization of the search population. The second is to complete the execution of the algorithm operators in the original CPA, and finally, to complete the execution of the introduced operators derived from the other two algorithms. The detailed pseudo-code of the presented algorithm can refer to Algorithm 1.

In this study, a kernel extreme learning machine driven by hybrid CPA for diagnosing COVID-19 from the perspective of biochemical indicators is proposed (ECPA-KELM). Figure 1 shows the flow diagram of ECPA-KELM. From this figure, it can be seen that the cross-validation mechanism is employed as the internal 5-and external 10-fold system, respectively, which means the 1 = 5 and 2 = 10 in this study. The detailed data segmentation diagram is as shown in Figure 2 .

The key part of this proposed method is the KELM, which uses the RBF kernel in terms of input space for mapping the aggregate data into a hidden layer space. The entire algorithm method covers the coefficient of penalty and kernel width , and the subset of features, the first penalty parameter sets the balance between minimizing fitting error and the complexity in the model, with the second kernel bandwidth defining nonlinear input spatial mapping into some high-dimensional space function.

The ECPA algorithm also develops these two parameters and synchronously converts the optimal feature subset, specifically, continuous space, into binary space utilizing the sigmoid function. The feature is considered to be se- 

If rand < g∕T = ; Calculate the by equation (1);

Calculate the by:

Calculate the by equation (6)-(8); Calculate the fitness of all Individuals; Update the ;

Calculate the by equation (9); Calculate the fitness of all Individuals; Update the ; = + 1;

End-Loop; lected if less than 0.5; otherwise, the characteristics would be discarded. Finally, the evolved KELM by ECPA gives an accurate early diagnosis of COVID-19 from the perspective of biochemical indicators.

The experimental section in this analysis consists of two parts. The first step will be to study the efficiency of the proposed ECPA, and the second part will use the proposed ECPA-KELM algorithm for a biochemical indicator diagnosis of COVID-19.

First of all, the efficiency of the proposed ECPA is extensively verified and carried out in comparison with other algorithms on IEEE CEC 2017 benchmark; these benchmarks are shown in Table 3 , and also strictly perforce the balance and diversity analysis of the improved ECPA and its original CPA. Several other algorithms, including CPA, CLPSO, DE, PSO, MFO, and GWO, were involved as competitors on the common benchmark. The algorithm parameters are specified concerning the original documents.

The following experimental examined extensively using the suggested ECPA algorithm to optimize a combination of the best parameter and KELM function subset, which is the result ECPA-KELM used for a biochemical indicator diagnosis of COVID-19 on data collection. In terms of ECPA-KELM, several common learning processes were also compared, including original KELM, GWO-KELM, MFO-KELM, PSO-KELM, SVM, and KNN. The two main parameters of [−2 ∧5 , 2 ∧5 ] and [−2 ∧5 , 2 ∧5 ], respectively, have been specified in the original KELM. The first parameter C can better off the minimization of fitting errors and the complexity of the model, the second parameter, , determines the nonlinear projection from the input area to a large space, and these two key parameters are particularly important to build the classification ability of KELM adapted to the current data set. In order to avoid the uncertainty in experiments caused by large data, before classification, data has been scaled to the range [-1, 1].

Notice that MATLAB simulation experiments were conducted on Windows Server 2018 R2 operator machine, the Xeon CPU E5-2660 v3 (2.60 GHz), and 16 GB of ram. We have charted our results based on fair comparison instructions and as per other works [85, 86, 87, 88] . A 10fold Cross-Validation (CV) is used to evaluate classification results to provide unbiased and objective results. Furthermore, four standard assessment parameters, including Specificity, Sensitivity, classification accuracy (ACC), and Matthews correlation coefficient are included (MCC), have been used for assessing the performance of ECPA-KELM. The detailed definition of the formula can refer to [89] . 6 . Experimental findings and analyses 6.1. Results of benchmark functions 6.1.1. The impact of GWO and MFO In this part, to estimate the effect of diverse mechanisms in ECPA and gain the best strategy combination, we experimented on IEEE CEC2017 30D benchmark tests [90] , in this test, each algorithm will be executed 30 times independently. In the algorithms, ECPA means both GWO and MFO are embedded in original CPA, GCPA indicates only GWO is introduced to basic CPA, MCPA shows only MFO is embedded into fundamental CPA. The results of Friedman's test on 30 functions are exhibited in Table 4 . From this table, it can be found that ECPA gains the lowest mean level value, 1.25235, and is the best; it signifies the combination between GWO and MFO outperforms single operator GWO or MFO, so the ECPA selected in the subsequent ex-Test set 1 Training set 1 Test set 1 Training set 1 

The benchmark of IEEE CEC2017 was used in this part to measure the property of the ECPA, and in several projects, these benchmarks have always been used [90, 91] . Furthermore, 30 separate experiments were carried out to mitigate the impact of random variables. In this analysis, the ECPA presented is contrasted with the CPA, CLPSO, DE, PSO, MFO, and GWO algorithms. Thirty individual executions carried out all these approaches to the CEC2017 standards.

In Table 5 , the average and standard deviation of the STD is shown to demonstrate detailed experimental results. The average ECPA results shown are the lowest among the benchmarks. Table 6 presents Friedman's ECPA test results against all other rivals. Following the average ranking of the algorithms concerned, the first best results of these benchmarks are disclosed in the ECPA, the worst findings being CPA, CLPSO, DE, PSO, MFO, and GWO. The main features of the current movement techniques abstracted from the grey wolf optimizer and moth-flame optimizer may be a reason for this. In this study, it can be obtained the original CPA between exploration and mining in this analysis.

The convergence curves of these involved algorithms on CEC2017 benchmarks to verify the performance of the designed ECPA are listed in Figure 3 . It can be observed from this figure that the designed ECPA shows the fast conference capabilities and obvious superiority to all other rivals in these CEC 2017 benchmarks. In addition, it also is noted that the designed ECPA has fast convergence searches such as F4, F7, and F9, which guarantee it to quickly obtain a theoretical optimal value. Furthermore, in terms of other benchmarks, the same convergence pattern is also observed. In short, the processes involved can be inferred that the property of the original CPA can significantly enhance. 

In this part, the proposed algorithm ECPA-KELM for diagnosing COVID-19 from the perspective of biochemical indicators is evaluated deeply. Table 7 shows the detailed results of ECPA-KELM on the collected COVID-19 data set. The 92.129% classification accuracy of the ECPA-KELM can be seen from this Table 7 , 90.506% of Matthew correlation coefficient, 92.298% of sensitivity, 89.627% of specificity, and their variance is 0.04379, 0.04379, 0.05322, and 0.06536 respectively. Furthermore, we can observe that the proposed ECPA-KELM can automatically acquire the optimum KELM model settings, mainly due to the enhanced ECPA, which can efficiently identify optimum settings and functions.

In addition, the methodology proposed ECPA-KELM is compared to original KELM and other evolutionary computingbased KELM, including CPA-KELM, original KELM, GWO-KELM, MFO-KELM, PSO-KELM, and two common algorithms, SVM and KNN, to check further the property of the ECPA-KELM model presented. Comparisons with accuracy, Matthew coefficient for correlation, susceptibility, specificity, and standard deviation are reported in a detailed statistical experiment in Table 8 and the comparative histogram of each experiment is also shown in Figure 4 in order to more visually represent the immediate difference in values.

The findings show that the ECPA-KELM algorithm is superior to other competitors and its corresponding standard deviation between all models is also less critical in four evaluation methods such as ACC, MCC, sensitivity, and specificity. In comparison to the original CPA-based KELM, the ECPA-KELM is more efficient and stable obviously. It should be noted that the original KELM, original SVM and KNN are all showing the worst performance for diagnosis of COVID-19 from the perspective of biochemical indicators, which can be preliminarily shown that KELM model selection capacity can be substantially improved by the algorithm proposed ECPA in this article and the ability to solve the accurate diagnosis of COVID-19 from the perspective of biochemical indicators. The second output of the KELM-based GWO-KELM is just under the ECPA-KELM, and the MFO-KELM and PSO-KELM perform very similar properties on this collected data. In this experiment, we can see that ECPA-KELM can automatically get the best property among all of these competing models, mostly because of the improved ECPA, where the optimal KELM parameters and the optimal subset of functions can be found automatically. Furthermore, the designed ECPA is used to perform parameter optimization and feature selection simultaneously for KELM to diagnose COVID-19 from the perspective of biochemical indicators. In this analysis, during the feature selection, the 10-fold CV method is used. The detailed selected amount of individual features and statistical values in each 10-fold cycle is shown in Table 9 . It can be observed that the ECPA-KELM proposed clearly exceeds others, and regarding the statistics, the features AGE, ALT, ALB, A/G, AST, and LDH were selected with values 9, 8, 9, 8, 9, and 8 respectively by the ECPA-KELM, while The other features have been comparatively picked with less. However, these features were not met by other rivals. Consequently, it can be inferred that such features, which often seem to be present, early recognition of COVID-19 and discrimination of other low-frequency features. Accordingly, due to the underlying details in these frequency features, further consideration should be provided in practice medical cases for these features of AGE, ALT, ALB, A/G, AST, and LDH.

In addition, the comparison results among these methods in terms of CPU time via 10-fold CV is recorded in Figure 5 . It can be observed that the original KELM consumes the least time and its execution speed is the fastest among all these algorithms, the original SVM takes the second least time. An explainable reason is that without the assistance of search algorithms, it will save a lot of algorithm execution time compared to those models-based search algorithms and the incidental result is that the classification performance of the algorithm is greatly reduced. It can also be noted that the designed ECPA-KELM consumes only the fourth-least time, which is more time than the original CPA-KELM, and it shows that the addition operator does increase the execution time of the algorithm. It is also worth noting that the time consumed by KNN and GWO-KELM to deal with this problem is close to the same and among all the algorithms, this PSO-KELM consumes the most time on this problem. A preliminary conclusion can be drawn that although the ECPA-KELM is not the least in terms of CPU consumption time, the four measurement values of it are the best, which also points us to a future research direction through reasonable parallel programming technique to achieve the reduction of CPU consumption time for ECPA-KELM.

In the present study, the diagnosis of COVID-19 from the perspective of biochemical indicators was investigated by using ECPA to perform KELM optimization parameters and feature selection simultaneously. Importantly, several key features were discovered, the features of AGE, ALT, ALB, A/G, AST, and LDH. Subsequently, an ECPA-KELM model is designed from the perspective of biochemical markers for an effective diagnosis of COVID-19. Thus, we think that the ECPA -KELM model will help to inform the decisionmaking process.

According to our observations, the proposed CPA-based method has shown enhanced exploratory and exploitative patterns to deal with more complex spaces can also, such as evaluation of human lower limb motions [92] , Lunar impact crater identification and age estimation [93] , shape registration [94] , regression tasks [95] , 3D deformable shape analysis [96, 97] , active surveillance [98] , service ecosystem [99, 100] , and micro-expression spotting [101, 102] . Also, we can test explorative features base on more classes of problems such as image editing [103, 104, 105] , engineering optimization problems [106, 107] , brain function prediction [108? ], epidemic prevention and control [109, 110] , large scale network analysis [111] , energy storage planning and scheduling [112] , image dehazing [113, 114, 115] , social recommendation and QOS-aware service composition [116, 117, 118] , medical diagnosis [119, 120, 121, 122] , covert communication system [123, 124? ] , pedestrian dead reckoning [125] , and feature selection [126, 127, 128] . Several studies have shown that age is one of the main risks of respiratory system diseases [129, 130] . In terms of SARS and MERS, older age was an independent predictor of SARS or MERS exacerbation risk and mortality [131, 132, 133] . Similarly, a large body of studies confirms that advanced age patients are more susceptible to COVID- Table 9 The numbers of selected feature

19 infections than young patients and older age patients are more susceptible to severe COVID-19 [134, 135] . Meanwhile, researchers have also shown that age is an independent pronouncing factor for COVID-19 [134, 136] . There are several possible reasons to explain this phenomenon. First, immune-senescence in aging is considered to be the leading cause of severe pneumonia mortality in older adults [129] . Second, with the increase of age, the cellular and humoral immune function of the body gradually declines [137, 138, 139] . For example, the level of immunoglobulin M and interferon decrease, the number of T-and B-lymphocyte decreases, resulting in an increased risk of infection [140] . Third, older COVID-19 patients tended to have more commodities, which was easier to acute respiratory failure and have a poor prognosis [141, 142] . Similar to their results, we found that the extreme COVID-19 group's mean age was 1.45 times higher than that of a non-severe COVID-19 group (P=0.00), indicating that age could be considered a promising clinical outcome index in COVID-19 patients. It was notified that approximately 60% of SARS patients have liver impairment [143] . Likewise, MERS patients also have liver damage [144] . Numerous retrospective studies have demonstrated that COVID-19 patients often have liver function damage. Based on the large retrospective data, increased levels of ALT and AST have been found in 14 to 53% of patients with COVID-19 [145, 146] . The most widely used parameters are ALT and AST liver functions. The permeability of the cell membrane will in- crease if hepatocytes are affected. The blood circulation is freed by high levels of cytoplasmic transaminases such as alanine aminotransferase, aspartate aminotransferase, and complete bilirubin [147, 148] . Many studies confirm that in extreme COVID-19 patients, ALT and AST were substantially higher than in non-severe patients [149] . However, the mechanism of SARS-CoV-2-induced liver impairment is not as yet clear. First, the cytokine storm following SARS-CoV-2 infection is thought to be one of the key factors of liver impairment [150] . Second, SARS-CoV-2 may directly be infecting hepatocytes. Xu and colleagues confirmed that the main pathologies of the liver in COVID-19 patients were characterized by moderate microvascular steatosis, mild lobular and portal activity [150] . In this analysis, we also found, relative to the non-extreme COVID-19 group, that the levels of ALT and AST in the severe COVID-19 group were 2.73 and 2.83 times higher. In summary, the association between liver function damage and COVID-19, a major factor in COVID-19 progression closely linked to COVID-19 seriousness, was revealed in these results. Another important biomarker, albumin (ALB), is one of the indexes of liver impairment [151] . ALB is synthesized by parenchymal cells in the liver, and the plasma half-life of albumin in the plasm is 15-19 days [152] . The level of ALB reflects the synthetic protein function of the liver, and ALB is a useful index for assessing nutritional status. He et al. found that plasma ALB levels are positively associated with the degree of community-acquired pneumonia (CAP) in pregnancy [153] . Recently, future analysis of over 400 CAP patients found a substantially higher plasma ALB level in the community of survivors than in the nonsurvivor group, indicating that ALB could be promising for CAP pronostics [154] . In line with these findings, Liu et al. found that plasma ALB levels in the group COVID-19 were considerably lower than in the group COVID-19 stabilization (36.62± 6.60 g/L vs. 41.27±4.55 g/L), suggesting hypoalbuminemia was positively associated with advanced COVID-19 progression and ALB may be used as an independent predictor of severity of illness and outcome [155, 156] . Consistent with their findings, in the current study, we also revealed that plasma ALB levels in the severe COVID-19 group were significantly lower than the non-severe COVID-19 group (35.90±4.65 vs. 41.53±2.57, P=0.000). In addition, globulin is the main component of serum non-albumin protein, which is composed of various pro-inflammatory proteins, such as immunoglobulin, complement, and C-reactive protein. Serum globulin levels are an objective marker to reflect the systemic inflammation and the immune status of the body [157, 158] . Of note, in recent years, Albumin/Globulin ratio (A/G) was commonly used to detect infectious diseases such as the acute exacerbation of the chronic obstructive pulmonary disease, hepatitis C and human immunodeficiency virus infection as a quick and inexpensive biomarker [159, 160, 161] . Furthermore, A/G also can be used as a novel predictor of prognosis in patients with a malignant tumor, including hepatocellular carcinoma, laryngeal squamous cell carcinoma, and colorectal cancer [162, 162] . However, few studies have reported on the relationship between A/G and COVID-19 patients. Universal research used by Zhou et al. shows that A/G was substantially related to COVID-19 severity. However, as shown by the multivariate binary logistic regression model, A/G was not an independent risk element for patients with COVID-19 [4] . Our analysis showed that in non-serious COVID-19 groups, A/G was substantially higher with approximately 1,41 times the amount of extreme COVID-19 groups (1.68±0.24 vs. 1.19± 0.36, P=0.000). This is the first time we realize that machine learning is being used to incorporate the A/G variable into COVID-19 research. All in all, the plasma ALB levels and The value for discrimination against COVID-19 patients A/G were shown to be significantly predictive and might predict COVID-19 progression.

Lactate dehydrogenase (LDH) is an essential energyproducing enzyme required for human physiology. LDH is present in almost all tissues, including liver, lung, kidney, skeletal muscle, myocardium tissue. Many studies have shown that elevated LDH levels are associated with disease progression and poor clinical outcome [163, 164] . For instance, a population-based study of 238 cases of SARS from Singapore suggested that high LDH is positive for adverse outcomes and acute syndrome of air distress [165] . Chang and colleagues reported that elevated LDH levels were related to increased mortality in SARS cases [166] . Therefore, LDH is a potential risk prediction factor. Recent seven studies, including a meta-analysis with > 1900 COVID-19 patients, found that the increased LDH level has been substantially linked to COVID-19 severity [167] . Our study found that the average LDH of the severe COVID-19 group was 382.95±152.78, and that of the non-severe COVID-19 group was 244.37±61.07, suggesting that plasma LDH levels may be regarded as a promising biomarker of clinical outcome in COVID-19 patients.

So far, very few relevant studies describing biochemical index, blood electrolyte, and clinical parameters to joint predict the severity and prognosis of the COVID-19. This is the first effort to incorporate age, ALT, AST, ALB, A/G, and LDH for predicting and discriminating COVID-19 severity using the machine learning method. However, some limitations exist in our research. First, the number of COVID-19 cases was relatively small, and patient data came from a single center. The model constructed ECPA-KELM can provide early warning for the severity of COVID-19 and help clinicians in the diagnosis and treatment of this infectious disease. In the future, we hope to enlarge the sample size and to improve the accuracy of the ECPA-KELM model further. Second, independent/ external datasets or prospective studies are needed to verify the accuracy of the ECPA-KELM model to make the model more reliable and stable.

The study uses clinical information from the Affiliated Yuqing Hospital of the Medical University of Wenzhou to develop an efficient ECPA-KELM early identification procedure and COVID-19 discrimination (Yueqing, China). The main innovation for the proposed methodology is for the current ECPA to include a new strategy to enhance and restore the original CPA search ability; the performance of the ECPA has been strictly regulated with the CEC2017 criteria compared with several other rivals. Experimental findings indicate that the ECPA proposed is much better suited than others to achieve this function optimization. In addition, ECPA has been proposed for the synchronized evolution of the optimum parameters and feature selection in KELM; the resulting ECPA-KELM was used successfully for early identification and discrimination against COVID-19. There has also been a rigorous analysis of the ECPA-KELM with other competitive algorithms. The findings also showed that the ECPA-KELM predicts the more stable properties more accurately and can be treated as a tool to provide early warning for the severity of COVID-19 and help clinicians in the diagnosis and treatment of this infectious disease.

For future work, a number of matters can be further investigated. More variables and coefficients are added, and parallel processing can also reduce the computing burden in the application phase; the following should be noted. We can also increase the number of data samples to create a safer and more effective prediction system. In addition, the proposed ECPA-KELM can also be employed to predict other variety of conditions such as clustering aspects and splitting the used image into CTs to expand the use of the developed system.

Beibei Shi and Hua Ye contributed equally to this paper and are co-first authors. Huiling Chen, and Peiliang Wu also contributed equally to this work and are co-corresponding authors. This research was supported by the Key Project of Zhejiang Provincial Natural Science Foundation under Grant (LD21F020001) and the National Natural Science Foundation of China (62076185, U1809209).

Early transmission dynamics in wuhan, china, of novel coronavirus-infected pneumonia

Classification of covid-19 patients from chest ct images using multi-objective differential evolution-based convolutional neural networks

Perspectives on mesenchymal stem/progenitor cells and their derivates as potential therapies for lung damage caused by covid-19

Prediction of the severity of corona virus disease 2019 and its adverse clinical outcomes

Suggestions for infection prevention and control in digestive endoscopy during current 2019-ncov pneumonia outbreak in wuhan, hubei province, china

Ppo-cpq: a privacypreserving optimization of clinical pathway query for e-healthcare systems

Artificial intelligence-based decision-making for age-related macular degeneration

Supervised machine learning based multi-task artificial intelligence classification of retinopathies

Predicting human microrna-disease associations based on support vector machine

Controllable synthesis of exceptionally small-sized superparamagnetic magnetite nanoparticles for ultrasensitive mr imaging and angiography

Large-scale one-pot synthesis of water-soluble and biocompatible upconversion nanoparticles for dual-modal imaging

Identifying medical diagnoses and treatable diseases by image-based deep learning

Automatic thyroid nodule recognition and diagnosis in ultrasound imaging with the yolov2 neural network

Predicting the future-big data, machine learning, and clinical medicine

Diagnosis of dementia by machine learning methods in epidemiological studies: a pilot exploratory study from south india. Social psychiatry and psychiatric epidemiology

Prediction model for health-related quality of life of elderly with chronic diseases using machine learning techniques

A review of dna-binding proteins prediction methods

Role of biological data mining and machine learning techniques in detecting and diagnosing the novel coronavirus (covid-19): a systematic review

Diagnosis of coronavirus disease 2019 (covid-19) with structured latent multi-view representation learning

A deep neural network to distinguish covid-19 from other chest diseases using x-ray images

Modified seir and ai prediction of the epidemics trend of covid-19 in china under public health interventions

Predicting covid-19 in china using hybrid ai model

Understanding the temporal evolution of covid-19 research through machine learning and natural language processing

Natural language processing and machine learning for detection of respiratory illness by chest ct imaging and tracking of covid-19 pandemic in the us

Covid-ct-md, covid-19 computed tomography scan dataset applicable in machine learning and deep learning

Hanadi Hassen, and Mohammad T Islam. An early warning tool for predicting mortality risk of covid-19 patients using machine learning

Using machine learning to develop a novel covid-19 vulnerability index (c19vi). Science of The Total Environment

Problem definitions and evaluation criteria for the cec 2017 competition on constrained real-parameter optimization

A scatter simulated annealing algorithm for the bi-objective scheduling problem for the wet station of semiconductor manufacturing

A modified particle swarm optimization algorithm for a batchprocessing machine scheduling problem with arbitrary release times and non-identical job sizes

Ant colony optimization with horizontal and vertical crossover search: Fundamental visions for multi-threshold image segmentation

Chaotic random spare ant colony optimization for multi-threshold image segmentation of 2d kapur entropy. Knowledge-Based Systems

Yingjie on Instrumentation Song, and Measurement. An improved quantuminspired differential evolution algorithm for deep belief network

Wu Deng, and measurement. Performance prediction using high-order differential mathematical morphology gradient spectrum entropy and extreme learning machine

Evolving an optimal kernel extreme learning machine by using an enhanced grey wolf optimization strategy

Sgoa: annealing-behaved grasshopper optimizer for global tasks. Engineering with Computers

Towards augmented kernel extreme learning models for bankruptcy prediction: Algorithmic behavior and comprehensive analysis

A two-layer nonlinear combination method for short-term wind speed prediction based on elm, enn, and lstm

Levy-based antlioninspired optimizers with orthogonal learning scheme

Harmonized salp chain-built optimization

Chaotic oppositional sine-cosine method for solving global optimization problems. Engineering with Computers

A multi-strategy enhanced salp swarm algorithm for global optimization

Modified extremal optimization for the hard maximum satisfiability problem

Zhengjiang Zhang, and Chongwei Backbone guided extremal optimization for the hard maximum satisfiability problem

Efficient boosted grey wolf optimizers for global search and kernel extreme learning machine training

Evolving support vector machines using fruit fly optimization for medical data classification. Knowledge-Based Systems

Chaotic multi-swarm whale optimizer boosted support vector machine for medical diagnosis

Toward an optimal kernel extreme learning machine using a chaotic moth-flame optimization strategy with applications in medical diagnoses

Design of fractional order pid controller for automatic regulator voltage system based on multi-objective extremal optimization

Binarycoded extremal optimization for the design of pid controllers

Adaptive population extremal optimization-based pid neural network for multivariable nonlinear control systems. Swarm and Evolutionary Computation

A novel gate resource allocation method using improved pso-based qea

An effective improved coevolution ant colony optimization algorithm with multi-strategies and its application

Orthogonal learning covariance matrix for defects of grey wolf optimizer: Insights, balance, diversity, and feature selection. Knowledge-Based Systems

An enhanced grey wolf optimization based feature selection wrapped kernel extreme learning machine for medical diagnosis

A fast approach for detection of erythemato-squamous diseases based on extreme learning machine with maximum relevance minimum redundancy feature selection

Gaussian mutational chaotic fruit fly-built optimization and feature selection

Boosted binary harris hawks optimizer and feature selection

A new machine-learning method to prognosticate paraquat poisoned patients by combining coagulation, liver, and kidney indices

A new fruit fly optimization algorithm enhanced support vector machine for diagnosis of breast cancer based on highlevel features

Developing a new intelligent system for the diagnosis of tuberculous pleural effusion. Computer methods and programs in biomedicine

Chaos enhanced grey wolf optimization wrapped elm for diagnosis of paraquat-poisoned patients

A two-stage feature selection method with its application

Feature selection based on improved ant colony optimization for online detection of foreign fiber in cotton

Predicting intentions of students for master programs using a chaos-induced sine cosine-based fuzzy k-nearest neighbor classifier

Predict the entrepreneurial intention of fresh graduate students based on an adaptive support vector machine framework

Predicting entrepreneurial intention of students: An extreme learning machine with gaussian barebone harris hawks optimizer

An improved grey wolf optimization strategy enhanced svm and its application in predicting the second major

Evaluation of sino foreign cooperative education project using orthogonal sine cosine optimized kernel extreme learning machine

Colony predation algorithm

Grey wolf optimizer: theory, literature review, and application in computational fluid dynamics problems

Moth-flame optimization algorithm: A novel nature-inspired heuristic paradigm. Knowledge-based systems

Extreme learning machine: theory and applications

Extreme learning machine for regression and multiclass classification

On a real-time scheduling problem

A bi-objective home healthcare routing and scheduling problem considering patients' satisfaction in a fuzzy environment

An improved jaya algorithm for solving the flexible job shop scheduling problem with transportation and setup times. Knowledge-Based Systems

Parameters estimation from the gravity anomaly caused by the two-dimensional horizontal thin sheet applying the global particle swarm algorithm

Hybrid genetic algorithm and a fuzzy logic classifier for heart disease diagnosis

Particle swarm optimization

A genetic algorithm tutorial

Grasshopper optimization algorithm for multi-objective optimization problems

Harris hawks optimization: Algorithm and applications. Future generation computer systems

Hunger games search: Visions, conception, implementation, deep analysis, perspectives, and towards performance shifts

Gene2vec: gene subsequence embedding for prediction of mammalian n6-methyladenosine sites from mrna

Deeplearning-enabled security issues in the internet of things

Fine-grained visual computing based on deep learning

Fuzzy quantized sampled-data control for extended dissipative analysis of t-s fuzzy system and its application to wpgss

Chaotic multi-swarm whale optimizer boosted support vector machine for medical diagnosis

Lshade with semi-parameter adaptation hybrid with cma-es for solving cec 2017 benchmark problems

Improved cuckoo search with better search capabilities for solving cec2017 benchmark problems

Using distributed wearable sensors to measure and evaluate human lower limb motions

Lunar impact crater identification and age estimation with chang'e data by deep and transfer learning

Topology-preserving nonlinear shape registration on the shape manifold

Supervised feature selection with orthogonal regression and feature weighting

Ferdous Sohel, and Hang Lei. Diffusion geometry derived keypoints and local descriptors for 3d deformable shape analysis

Scale space clustering evolution for salient region detection on 3d deformable shapes

Active surveillance via group sparse bayesian learning

Value entropy: A systematic evaluation model of service ecosystem evolution

Social learning evolution (sle): Computational experiment-based modeling framework of social manufacturing

Local temporal pattern and data augmentation for micro-expression spotting

Mesnet: A convolutional neural network for spotting multi-scale microexpression intervals in long videos

Xiaogang Multimedia Tools Jin, and Applications. Semantic portrait color transfer with internet images. Multimedia Tools and Applications

Xiaoyang Mao, and Junru Parallel and efficient approximate nearest patch matching for image editing applications

Kun transactions on visualization Zhou, and computer graphics. Parallel style-aware image cloning for artworks

Levy-based antlioninspired optimizers with orthogonal learning scheme

Chaotic oppositional sine-cosine method for solving global optimization problems. Engineering with Computers

Prediction of trust propensity from intrinsic brain morphology and functional connectome

Mining spatiotemporal diffusion network: A new framework of active surveillance planning

Next generation technology for epidemic prevention and control: Data-driven contact tracking

A scalable redefined stochastic blockmodel

Riskaverse storage planning for improving res hosting capacity under uncertain siting choice

Self-filtering image dehazing with self-supporting module

Haze concentration adaptive network for image dehazing

Pyramid channel-based feature attention network for image dehazing. Computer Vision and Image Understanding

Towards context-aware social recommendation via individual trust. Knowledge-Based Systems

A probability distribution detection based hybrid ensemble qos prediction approach

An efficient and reliable approach for quality-ofservice-aware service composition

Diagnosis of alzheimer's disease based on deeply-fused nets

Projective parameter transfer based sparse multiple empirical kernel learning machine for diagnosis of brain disease

Uncertainty modeling for multi center autism spectrum disorder classification using takagi-sugenokang fuzzy systems

A novel deep-learning model for automatic detection and classification of breast cancer using the transfer-learning technique

Research on a covert communication model realized by using smart contracts in blockchain environment

A covert communication method using special bitcoin addresses generated by vanitygen

Inertial/magnetic sensors based pedestrian dead reckoning by means of multi-sensor fusion

Adaptive data structure regularized multiclass discriminative feature selection

Top-k feature selection framework using robust 0-1 integer programming

Robust hand tracking via novel multi-cue integration

Pneumonia in the elderly: a review of the epidemiology, pathogenesis, microbiology, and clinical features

Pneumonia in the elderly: understanding the characteristics

Mortality risk factors for middle east respiratory syndrome outbreak, south korea

Predictors of mortality in middle east respiratory syndrome (mers)

Outcomes and prognostic factors in 267 patients with severe acute respiratory syndrome in hong kong

Clinical characteristics of 138 hospitalized patients with 2019 novel coronavirus-infected pneumonia in wuhan, china

Clinical course and risk factors for mortality of adult inpatients with covid-19 in wuhan, china: a retrospective cohort study. The lancet

Estimating clinical severity of covid-19 from the transmission dynamics in wuhan, china

Changes in the b-cell repertoire with age

The aging of the immune system

The immunopathogenesis of sepsis in elderly patients

Effects of influenza immunization on pneumonia in the elderly

Clinical characteristics of novel coronavirus cases in tertiary hospitals in hubei province

Early detection and disease assessment of patients with novel coronavirus pneumonia. Zhonghua jie he he hu xi za zhi= Zhonghua Jiehe he Huxi Zazhi= Chinese Journal of Tuberculosis and Respiratory Diseases

Sars-associated viral hepatitis caused by a novel coronavirus: report of three cases

Histopathology of middle east respiratory syndrome coronovirus (mers-cov) infection-clinicopathological and ultrastructural study

Radiological findings from 81 patients with covid-19 pneumonia in wuhan, china: a descriptive study

Epidemiological and clinical characteristics of 99 cases of 2019 novel coronavirus pneumonia in wuhan, china: a descriptive study

Protective mechanism of glycyrrhizin on acute liver injury induced by carbon tetrachloride in mice

Allergenicity attributes of different peanut market types

Clinical features of patients infected with 2019 novel coronavirus in wuhan, china. The lancet

Pathological findings of covid-19 associated with acute respiratory distress syndrome. The Lancet respiratory medicine

Quillaja saponaria bark saponin protects wistar rats against ferrous sulphate-induced oxidative and inflammatory liver damage

Effect of early enteral nutrition on patients with digestive tract surgery: A meta-analysis of randomized controlled trials

Anemia and low albumin levels are associated with severe community-acquired pneumonia in pregnancy: a case-control study

Tae Youn Kim, Sang Hoon Na, and Seung Sik Hwang. Albumin and c-reactive protein have prognostic significance in patients with community-acquired pneumonia

Analysis of factors associated with disease outcomes in hospitalized patients with 2019 novel coronavirus disease. Chinese medical journal

Risk factors for disease severity, unimprovement, and mortality in covid-19 patients in wuhan, china. Clinical Microbiology and Infection

Baseline serum globulin as a predictor of the recurrence of lone atrial fibrillation after radiofrequency catheter ablation

The pretreatment albumin to globulin ratio has predictive value for long-term mortality in nasopharyngeal carcinoma

Application of albumin/globulin ratio in elderly patients with acute exacerbation of chronic obstructive pulmonary disease

Effect of combination antiretroviral treatment on total protein and calculated globulin levels among hiv-infected patients

Peripheral tnf elevations in abstinent alcoholics are associated with hepatitis c infection

Preoperative albumin/globulin ratio has predictive value for patients with laryngeal squamous cell carcinoma

Prognostic factors of early fatal outcome and long-term survival in patients with pneumocystis carinii pneumonia and acquired immunodeficiency syndrome

Multisystem organ failure predicts mortality of icu patients with acute respiratory failure secondary to aids-related pcp

Sars in singapore-predictors of disease severity

Hematological and biochemical factors predicting sars fatality in taiwan

Laboratory findings of covid-19: a systematic review and metaanalysis. Scandinavian journal of clinical and laboratory investigation

J o u r n a l P r e -p r o o f (1) The ECPA-KELM is designed to diagnose COVID-19 from the perspective of biochemical indicators.(2) Performance of the CPA is enhanced special operators from other algorithms.(3) Property of the ECPA is verified on CEC2017 optimization tasks. 

We wish to draw the attention of the Editor to the following facts which may be considered as potential conflicts of interest and to significant financial contributions to this work.We confirm that the manuscript has been read and approved by all named authors and that there are no other persons who satisfied the criteria for authorship but are not listed. We further confirm that the order of authors listed in the manuscript has been approved by all of us.We confirm that we have given due consideration to the protection of intellectual property associated with this work and that there are no impediments to publication, including the timing of publication, with respect to intellectual property. In so doing we confirm that we have followed the regulations of our institutions concerning intellectual property.We understand that the Corresponding Author is the sole contact for the Editorial process (including Editorial Manager and direct communications with the office). He/she is responsible for communicating with the other authors about progress, submissions of revisions and final approval of proofs. We confirm that we have provided a current, correct email address which is accessible by the Corresponding Author and which has been configured to accept email from Huiling Chen(chenhuiling.jlu@gmail.com) Signed by all authors as follows: