Impact analysis of recovery cases due to COVID19 using LSTM deep learning model

1 Introduction

Now a days, COVID-19 is a global pandemic to the health issue word-wide. Not only, to combat to the unknown disease is a greatest challenge. Broadly speaking, human are affected by COVID-19. This virus belongs to Nidovirales which consist of Arteriviridae, Roniviridae, and Coronaviridae [24, 25]. In the meantime, the family of the Coronaviridae is divided into Coronavirinae and Torovirinae. Furthermore, it is categorized into four types of variant’s such as gamma, delta, alpha and beta. It is one kind of RNA genome virus. The dimension of this virus is 26 to 32 kilobases and visible by the electron microscope. World Health Organization (WHO) conducted several research on COVID-19 to predict the new strain of COV named as COVID-19. The peoples are getting infected more and the illness is spreading tremendously word-wide. Therefore, the word is facing threat due to the pandemic of COVID-19 [26]

COVID-19 was started in Wuhan city, capital of Hubei provenance in China at the end of 2019 December while Chinese peoples are getting preparation to celebrate their spring festival[1]. For the outbreak of Covid 19 later it was named as novel coronavirus (Covid-19) by the WHO(Word Health Organization). Hundred of thousand infected people left Wuhan which is the central hub of the provenance and they are identified as first transmission media to carry this virus. The virus SARS-CoV2, causing the Coronavirus disease (COVID-19), has spread from Wuhan, China to every populated continent[2]. As the COVID-19 outbreak is now a pandemic, it will be important to have tools to rapidly identify those at risk of morbidity and mortality. We must augment clinical skills to identify mild cases that will progress to critical illness[3].

Predictive analysis (a form of artificial intelligence) learns from historical data to help forecasting future outcomes. This analysis uses machine learning algorithms that can extract insights and rules from recorded data with the most predictive tool for making accurate predictions. The analysis results in predicting the tentative future contexts that can be conducive. At present, its the demands of time what will be the scenario of COVID-19 in future. Everyday people’s are getting sick due to pandemic of COVID-19 and word is facing a public health threat

[26]. In this situation, the prevent of COVID-19 is an emergency needed. By the way, still now there is no exact cure. In such situation, machine learning approach is a viable solution to find the exact battle to fight against the COVID-19. Moreover, the impact of recovery cases is a vital issue. Many works have already been done using machine learning. In [27] dynamics of susceptible, infected and recovered cases are studied. All the cases are simulated considering both on and off immigration condition. All the plots here are shown from mathematical model. So there may be discrepancies from the practical data. The work in [28] Covid-19 outbreak has been modeled mathematically by introducing Caputo Fractional derivatives. In [29] a compartmental model was made to formulate the epidemic. The effect of vaccination was also deemed.

The main motivation of this work come from the analysis of the death across the world using AI based technology. The objective of this work is to predict the impact of recovery cases using LSTM deep learning model. The job done here are:

Development of a python code by Pytorch
Collection of data as CSV file from authentic source from internet.
Using the code and data the prognosis of recovery number in future days was performed.

2 Related Work

The virus SARS-CoV2, which causes coronavirus disease (COVID-19) has scattered as a pandemic form and has spread to every continent. We must augment clinical skills to identify mild cases that will progress to critical illness. The study in[4] showed that the outbreak seen has not matched the clinical severity seen in the initial Wuhan, Hubei epicenter. A clear limitation of this study is the size of the data set though the models were 70% - 80% predictive.

In [5], by the utilization of COVID-19 daily data of mainland China excluding Hubei province, comprised of the cumulative confirmed cases, the cumulative deaths, newly confirmed cases and the cumulative recovered cases.

The aim of [6] was to assess the machine and deep learning-based architectures performance that is proposed in the present year for classification of Coronavirus images such as, X-Ray and computed tomography. The target of the study in [7]

was to estimate the basic reproduction number of the Wuhan novel coronavirus using SEIR compartment model to characterize the early spreading of 2019-nCoV. In

[8], Logistic model, Bertalanffy model and Gompertz model, which are relatively simple but support the statistical law of epidemiology, are selected to predict the epidemic situation of COVID-19.

In [9]

, the analysis of the cumulative confirmed cases, cumulative deaths, and cumulative cured cases was conducted using an Elman neural network, long short-term memory (LSTM), and support vector machine (SVM).

In the work of [10], the COVID-19 outbreak in India has been predicted based on the pattern of China using a machine learning approach. This outbreak is predicted between the third and fourth weeks of April 2020 in India to be controlled around the end of May 2020.

Taking into account the uncertainty due to the incomplete data of infective population, they applied the well-known SEIR compartmental model for the prediction was found in [11].

In the study of [12] COVID-19 daily new cases and deaths in the USA from two population-based data-sets were extracted. Study of [13] demonstrates that real-time, publicly available ensemble forecasts issued in April-July 2020 provided robust short-term predictions of reported COVID-19 deaths in the United States. The paper of [14] introduces an objective approach to predicting the continuation of the COVID-19 using a simple, but powerful method to do so.

The propose of [15] was a shallow long short-term memory (LSTM) based neural network to predict the risk category of a country. A Bayesian optimization framework have been used to optimize and automatically design country-specific networks. In COVID-19 detection, a modification version of U-net network was applied to identify the covid and non-Covid case classification purpose. In this method, X-ray flim image data was considered [16]. On the other hand, CT image based COVID-19 classification was done in [17]. VGG16, Res-Net, CNN, auto encoder neural network networks were applied to compare the computational accuracy. The spreading case of COVID-19 was analyzed in [18] for high rise countries. Mutimodal based emotional model was proposed to analyze COVID19 by Author Gupta in [19]. A threshold condition was studied in [20] for COVID19 disease free toward the global sustainability.

Alok Negi et. al. worked on face mask detection in public places [30][31]. Malaria infection detection was done in [33]. Multi-view video summarizing was done in [32]. In [34]Hidden Markov Model(HMM) was utilized to detect the depression of mind. The thoracic surgery can be assisted with AI which was suggested in [35]. In [36][37] plant disease was simulated by AI. Face mask from video streaming was performed in [38].A human action recognition system was designed by RESNET50 that can determine the human daily activities[39]. In [40]

liver disease detection was proposed to reduce the instigation of death. Unlike the SVM(Support Vector Machine), ANN(Artificial Neural Network) technique, the CNN was applied for classify the cancer cell into Benign and Malignant in

[41].

Qingzhen Xu[42] showed that financial prediction can be done using machine learning. In [43]

the saliency map of a complex image is proposed. Osteoporosis is predicted with transfer learning in

[44]. Trend of stock is predicted in [45]. Drosophila embryonic gene expression computation requires accuracy and rapidity that is presented in [46].

Reference

Advantage

Limitation

Jiang et al. [7]

Its identify the combinations

of clinical characteristics of

COVID-19 that predict outcomes.

Besides, a tool with AI capabilities

that will predict patients at risk for

more severe illness on initial

presentation has developed.

Limited amount of data was used.

Also, no use of update technology

AI tool such as LSTM,

Efficient-Net and others.

Azimi et al.[8]

Estimate the reproduction of

diseases affected. Exponential

growth rate, maximum likelihood

was evaluated.

The estimation is based on

study cases not on real case data.

In addition, simulated daily cases

are used lead the missing of real

result due to sampling information

of PCR.

Salehi et al.

[9]

Its provides art of state for

future of AI technology.

Also, X-ray and CT scan

data set were discussed.

Draw only low amount of data.

Discuss only old model of

AI technique.

Zhou et al.

[10]

Estimate the basic reproduction

of the confirm and suspected cases.

Data acquire from severe acute

respiratory syndrome.

Intervention was not for

propagation analysis.

Timely and effective

control measures was absent for

further transmission analysis.

Hao et al. [11]

Various kinds of AI based

network are used to predict

new cured cases, confirmed

cases and new deaths cases.

The proposed model is less

robust since the average value

is larger than actual. Further

performance improvement is

required

Tiwari et. al.

[12]

Time series forecasting method

to forecast the death case,

confirm case and recovery case.

Unclear about which time series

method is used for forecasting.

Also, did not consider

socioeconomic consideration.

Samuel et al.

[13]

Various model were used to

validate the existing

mathematical model.

No use of AI model for prediction.

Kuniya et al.

[14]

Estimation method is used for

prediction of transmission of

reproduction.

Use to determine only

basic reproduction of

Covid19 affection.

Yuan et. al.

[15]

In this work an internet search

based model was developed.

This model seemed to be useful

because there is a correlation

between the trends in daily new

cases and in new deaths and the

internet search terms.

In spite of the correlation between

the search terms and trends of

death cases, the accuracy was

very poor. Thus this

model should be further

developed.

Ray et. al.[16]

In this work ensemble model

was introduced. Here multiple

model prediction was incorporated

into a combined forecast.

Ensemble model distill

information of multiple forecast.

Ensemble forecast weighted

models equally which should

be done deeming the record of

performance. To remove the

uncertainty and raise the precision

longer horizon can be considered.

Petropoulos

et. al.[17]

In this exercise statistical forecast

was provided using robust time

series model. The government

and individual can take steps

using the forecast.

Here an assumption that future

will follow the past was made.

But this may not be true always.

Pal et. al.[18]

This work introduced the use of

shallow LSTM. This model

out-performs the state-of-the-art

models.

Here weather data was combined

for prediction. Various modalities

of data can be combined like

flight, business, travellers etc.

Rahman

et. al.[19]

In this work, a large data set

of chest X-Ray was involved.

Image enhancement and

segmentationwere done which

were unprecedented. The standard

U-net was modified

which yielded a nice accuracy in

detection. This accuracy helped

detecting the spread of the virus.

Though this work is a benchmark,

it arises complexity.

Fouladi

et.al.[20]

Classification of Covid-19 CT

images was performed using

modified VGG-16, ResNet-50 etc.

DL models. The modifications of

parameters and hyper parameters

introduce us with new models

for DL. The new models can be

utilized to classify Ebola,

SARS etc. images.

Data augmentation was not

performed here. Larger data set

can be entailed. Quantity of

infections should also be measured.

Mahmoudi

et. el[21]

The relationship between spread of

Covid-19 and population’s size

was studied. This seems to be

necessary for policy makers.

Fuzzy clustering was used

here. But regression model,

time series models and artificial

intelligence model could be

used here.

Alok Negi

et. al.[27]

Here using VGG16 and CNN on a

Simulated Masked Face Dataset

(SMFD), precautionary measures

are suggested.

The model was not used for video

surveillance datasets.

Alok Negi

et. al.[28]

Using Keras-surgeon, model pruning

was done to simplify the neural

network. This model can be used to

detect mask wearing faces in public

places

Comparison between Keras and

Pytorch was not presented.

Alok Negi

et. al. [29]

To detect malaria infection CNN

model was developed. Result was

very satisfactory.

Comparison between Keras and

Pytorch was not present.

Krishnan

Kumar et. al.

[30]

Facilitates the summarization of

multi-view video.

Mono-view video summarization

is not possible here.

Haroon

Ansari et. al.

[31]

Using Hidden Markov Model the

depression of human mind is

detected.

The procedure may yield result

as depression in place of

exhilaration.

Anshuman

Darbari et. al.

[32]

Encourages the use of AI in the

field of thoracic surgery.

Practical thoracic surgery may not

not be feasible by AI

Alok Negi

et. al.[33]

Various plant diseases were detected

using CNN

This deep learning model can be

used for only one type of plant

classification.

Alok Negi

et. al. [34]

The diseases of citrus fruit and

leaves were detected using CNN.

This was a good job to detect the

defected plants and take proper

medical treatment.

This defect detection can be

compared with other deep learning

models.

Qingzhen Xu.

[39]

A machine learning method is

improved by a two dimensional

equation is found by finite difference

method. This method can be used

for financial prediction.

This technique was not applied in

the field of science and technology.

Qingzhen Xu

et.al.[40]

For analyzing the complex images,

saliency detection model is produced

through a new method.

Comparison among various

methods is not presented.

Qingzhen Xu

et. al.[41]

Fall of osteoporosis is predicted by

transfer learning with s high

accuracy.

Comparison among various

methods is not presented.

Qingzhen Xu

et.al.[42]

Prediction of trend of stock based on

money flow using fuzzy clustering

and collaborative filtering can be

possible from this work

Considering the investor

characteristics this model can be

improved.

Table 1: Related Work In a Nutshell

Below table shows the comparison of the proposed method compare to others.

Model

Advantage

Limitation

LSTM

1. Capable of learning long-range dependencies.

2. Able to solve dependencies problem.

3. Work as a memory network with large dataset.

1. Complex architecture.

2. Repeating module has complex layers.

3. Unable to remove vanishing gradients issue completely

GRU

1. Use less memory compared to LSTM approach

2. Faster training and execution capability.

1. Able to train less parameters for training purpose.

2. Provides lower accuracy.

Linear Regression

1. Provides better result when data sets are linear.

2. Using dimensional reduction over-fitting can be avoided.

1. Under-fitting could not be mitigated.

2. Lower accuracy with poor performance.

Table 2: Comparison among Deep Learning Models

3 Proposed Methodology

In this section, the methodology of the research proposed are divided into three parts. The data collection for this work is describe in section 3.1. Section 3.2 describe the MinMax Scalar method to format the collected data. Then prepare the data for LSTM deep learning model for training and testing purpose. The details of LSTM architecture is given in section 3.3.
At first the data is read as data frame from CSV file. The graph of cumulative death per day and change of death per day are plotted. The data is partitioned as training and testing data sets. A predictive model is constructed by LSTM. Using this model and training data, anticipation is accomplished. The prediction and testing data are compared and checked for validation of the proposed model. Later, the whole data is used in the model. Again the validation step is performed. Finally, for comparative exhibition of historical and predictive daily cases, a figure is plotted.

3.1 Data Collection and Description

An updated recovery cases data set were collected from Github repository which automatically update the Covid-19 data in daily cases. The time series data have sorted for prediction purpose. The data of the repository handled by the CSSE(Central for System Science and Engineering) of Johns Hopkins University. The data-sets contain the recovery cases across global as a csv data file between 22 January, 2020 to 27 February, 2021 [21]. The data-set contains 258 regions recovery cases per day due to Covid-19 pandemic.

Figure 1: Recovery cases for various countries. The countries are alphabetically arranged from Afghanistan to Cyprus

Figure 2: Recovery cases for various countries. The countries are alphabetically arranged from Djibouti to Moldova

Figure 3: Recovery cases for various countries. The countries are alphabetically arranged from Namibia to Zimbabwe

Figure 1,2 ,3

shows the overview of frequent recovery happens in countries using Raw data. This picture indicates that India is the highest recovering country wheres Brazil is second in recovering of life. Other countries such as U.S., Russia exhibit moderate amount of recovery. In every moment people are migrating from each country to others. Therefore, the countries should take proper steps to stop the infected people migration from neighbouring countries. Otherwise, the nearest country will have to pay high death toll.

3.2 Feature Scaling

The machine learning algorithm do not perform well with different scale data. To avoid this issue, the min-max features scaling transform perform well to scale the data in the same platform [22]. Sometime, its known as normalization since its normalize the between 0 to 1. Its operation can be expressed using the following procedure:

X_{s c} = \frac{X - X_{m i n}}{X_{m a x} - X_{m i n}} .

(1)

Where $X$ is the original data-set. $X_{m i n}$ is the minimum value in the data range whereas $X_{m a x}$ determine the maximum value of the data.

3.3 Architecture of Long Short Term Memory (LSTM)

LSTM (Long Short Term Memory) is mainly approached to predict the future tread trend of confirm death case using history of the tanning data. LSTM is the one kind of RNN network. Generally, RNN works on short-term memory basis wheres LSTM on short and long-term memory. The major drawback of RNN is the disappearance of gradients descent. LSTM address this issue by combining the long-term memory with short-term through the subtle control gate. In 1997, LSTM was proposed by Hochreiter and Schmidhuber [23]. Figure 4 shows the general architecture of the LSTM model. Three gates are basically used named as input gate, forget gate and output gate. Initially, the input data is feed to the forget through input gate. The forget gate decide which information is discard by reading the hidden state, $h_{t - 1}$ and input layer $x_{t}$ information.It generates an output vector between zero and 1. This value of vector determine how much amount of information is discarded or retained in the state cell $C_{t}$ .The zero value of the vector indicates all information is discarded and one means it retain all information.

After that, there is two steps to identify how much information is added to the cell. First, through an input gate operation $h_{t - 1}$ , and $x_{t}$ decide which information to update.

z_{f} = σ (W_{f} . [h_{t - 1}, x_{t}] + b_{f}) .

(2)

z_{i} = σ (W_{i} . [h_{t - 1}, x_{t}] + b_{i}) .

(3)

The new candidate cell $Z$ is acquired through a $t a n h$ layer using $h_{t - 1}$ , and $x_{t}$ .

z = t a n h (W . [h_{t - 1}, x_{t}] + b) .

(4)

In the later stage, the cell is updated as follows:

c_{t} = z_{f} * C_{t - 1} + z_{i} * z .

(5)

Where * and + represent vector multiplication and addition. At last, the output of the cell should be identified after updating. This is done via the help of $h_{t - 1}$ , and $x_{t}$

. An Sigmoid activation function output gate is applied to mapping the output. Than, to obtain a vector between +1 to -1, the cell state pass through the

t a n h

layer.

z_{0} = σ (W_{o} . [h_{t - 1}, x_{t}] + b_{0}) .

(6)

h_{t} = z_{o} * t a n h (c_{t}) .

(7)

where $z_{f}$ , $z_{o}$ and $z_{i}$ are the gate control state of the forget gate, output gate, and input gate, respectively. Also, z is known as candidate input cell through the layer of $t a n h$ .

z_{f} = σ (W_{f} . [h_{t - 1}, x_{t}] + b_{f}) .

(8)

z_{i} = σ (W_{i} . [h_{t - 1}, x_{t}] + b_{i}) .

(9)

4 Results and Discussion

This section presents all computational results related to this research work.
Figure 5 shows the graphical visualization of changing in the number of recovery cases for 403 days. In this figure, the cumulative recovery case of 22 January, 2020 used as a reference. This value is deduced from the cumulative deaths of other days. Thus from this figure we find that at some days the number of death increases and at other days the number decreases.

Figure 5: Daily recovery cases across global according to real data.

The cumulative case of everyday recovery case is figured out in Figure 6. In this case, the numbers of recovery per day from 22 January 2020 to 01 January, 2021 of 271 regions in the world are plotted. From this figure, it is concluded that, the number of recovery is increased most of the days.

Figure 6: Cumulative case of daily recovery across globe.

The comparative study of the historical, real and predicted results are presented in Figure 7. This figure tells that the historical daily cases obtained from 379 days of training data and the rest of 24 days data used as a real case. Based on historical data, the prediction is done using LSTM approach. For the 24 days the number of predicted recoveries decreases slightly compared to real cases.

Figure 7: Combined show of Historical, Real and Predicted daily recovery cases.

Figure 8

shows the prediction result of daily cases using whole data sets. The prediction is for 20 days with 60 number of epochs and 24 batch size. In this cases, the predicted daily cases decaying fast until 03 March 2021 and than obtain an almost stable position till 19 March 2021.

The historical and predicted daily cases is shown Figure 9 with considering whole data sets. The number of prediction is decrease. This may be happened due to majority of days provides falling recovery history.

Figure 9: Combined show of historical and predicted daily recovery cases.

5 Conclusions

There is a contingency of spark of this global issue. Still many people are neglecting the panic. So we must need the proper decision. Forecasting and uncertainty are the integral part of decision making. Tentative anticipation of future is not only a concern of health sector but also of politics and economy. Our study is beneficial for this prediction. Our one can be used to combat against the second wave. This algorithm can be more affluent if more update data is available. Lacking of our anticipation emerges from demography, hospital capacity and social distances.

Declarations

The following sections contain the information regarding the declaration.

Funding: Not Applicable
Conflict of interest/Competing interests:The authors declare that there is no conflict of interest.
Ethics approval:This work has not been published anywhere yet.
Consent to participate:Everyone among us opined to participate in this study.
Consent for publication:Everyone of us is agreed to publish the article in this journal.
Availability of data and materials:Not Applicable.
Code availability:The code is not available in the cloud.
Authors’ contributions:All the authors have equal contribution in this article.

References

[1] Acosta-Quiroz, Johana and Iglesias-Osores, Sebastián. Salud mental en trabajadores expuestos a COVID-19. Revista de Neuro-Psiquiatría, volume.83, number.3,pages.212–213, 2020,Universidad Peruana Cayetano Heredia. Facultad de Medicina
[2] Waibler, Zoe and Anzaghe, Martina and Ludwig, Holger and Akira, Shizuo and Weiss, Siegfried and Sutter, Gerd and Kalinke, Ulrich. Modified vaccinia virus Ankara induces Toll-like receptor-independent type I interferon responses. Journal of virology, volume.81, number.22, pages.12102–12110, 2007, Am Soc Microbiol.
[3] Rath, Smita and Tripathy, Alakananda and Tripathy, Alok Ranjan. Prediction of new active cases of coronavirus disease (COVID-19) pandemic using multiple linear regression model. Diabetes & Metabolic Syndrome: Clinical Research & Reviews, volume.14, number.5, pages.1467–1474, 2020, Elsevier
[4] Jiang, Xiangao and Coffee, Megan and Bari, Anasse and Wang, Junzhang and Jiang, Xinyue and Huang, Jianping and Shi, Jichan and Dai, Jianyi and Cai, Jing and Zhang, Tianxiao and others. Towards an artificial intelligence framework for data-driven prediction of coronavirus clinical severity. Computers, Materials & Continua, volume.63, number.1, pages.537–551, 2020
[5] Wan, Hui and Cui, Jing-an and Yang, Guo-jing. Risk estimation and prediction by modeling the transmission of the novel coronavirus (COVID-19) in mainland China excluding Hubei province. medRxiv, 2020,Cold Spring Harbor Laboratory Press.
[6] Salehi, Ahmad Waleed and Baglat, Preety and Gupta, Gaurav. Review on machine and deep learning models for the detection and prediction of Coronavirus. Materials Today: Proceedings, volume.63, number.1, pages.537–551, .2020 pages.3896–3901,.2020,Elsevier.
[7] Zhou, Tao and Liu, Quanhui and Yang, Zimo and Liao, Jingyi and Yang, Kexin and Bai, Wei and Lu, Xin and Zhang, Wei. Preliminary prediction of the basic reproduction number of the Wuhan novel coronavirus 2019-nCoV. Journal of Evidence-Based Medicine, pages.3–7, 2020, Wiley Online Library.
[8] Acosta-Quiroz, Johana and Iglesias-Osores, Sebastián. Salud mental en trabajadores expuestos a COVID-19. Revista de Neuro-Psiquiatría, volume.83, number.3, pages.212–213, 2020, Universidad Peruana Cayetano Heredia. Facultad de Medicina.
[9] Hao, Yan and Xu, Ting and Hu, Hongping and Wang, Peng and Bai, Yanping. Prediction and analysis of corona virus disease 2019. Revista PloS one, volume.15,number.10, pages.e0239960, 2020 .Public Library of Science San Francisco, CA USA.
[10] Tiwari, Sunita and Kumar, Sushil and Guleria, Kalpna. Outbreak trends of coronavirus disease–2019 in India: a prediction. Disaster medicine and public health preparedness, volume.14,number.5, pages.e33–e38, 2020,Cambridge University Press.
[11] Kuniya, Toshikazu. Prediction of the epidemic peak of coronavirus disease in Japan, 2020. Journal of clinical medicine, volume.9,number.3, pages.789, 2020, Multidisciplinary Digital Publishing Institute.
[12] Yuan, Xiaoling and Xu, Jie and Hussain, Sabiha and Wang, He and Gao, Nan and Zhang, Lanjing. Trends and prediction in daily new cases and deaths of COVID-19 in the United States: an internet search-interest based model. Exploratory research and hypothesis in medicine, volume.5, number.2, pages.1, 2020
[13] Ray, Evan L and Wattanachit, Nutcha and Niemi, Jarad and Kanji, Abdul Hannan and House, Katie and Cramer, Estee Y and Bracher, Johannes and Zheng, Andrew and Yamana, Teresa K and Xiong, Xinyue and others. Ensemble forecasts of coronavirus disease 2019 (COVID-19) in the US, MedRXiv, ]2020, ]Cold Spring Harbor Laboratory Press
[14] Petropoulos, Fotios and Makridakis, Spyros, Forecasting the novel coronavirus COVID-19. PloS one, volume.15, number.3, pages.e0231236, 2020, Public Library of Science San Francisco, CA USA.
[15] Pal, Ratnabali and Sekh, Arif Ahmed and Kar, Samarjit and Prasad, Dilip K. Neural network based country wise risk prediction of COVID-19. Applied Sciences, volume.10, number.18,pages.6448, 2020, Multidisciplinary Digital Publishing Institute.
[16] Rahman, Tawsifur and Khandakar, Amith and Qiblawey, Yazan and Tahir, Anas and Kiranyaz, Serkan and Kashem, Saad Bin Abul and Islam, Mohammad Tariqul and Al Maadeed, Somaya and Zughaier, Susu M and Khan, Muhammad Salman and others. Exploring the effect of image enhancement techniques on COVID-19 detection using chest X-ray images, Computers in biology and medicine, volume.132, pages.104319, 2021, Elsevier
[17] Fouladi, Saman and Ebadi, MJ and Safaei, Ali A and Bajuri, Mohd Yazid and Ahmadian, Ali. Efficient deep neural networks for classification of COVID-19 based on CT images: Virtualization via software defined radio. Computer Communications, 2021,Elsevier
[18] , Mahmoudi, Mohammad Reza and Baleanu, Dumitru and Mansor, Zulkefli and Tuan, Bui Anh and Pho, Kim-Hung. Fuzzy clustering method to compare the spread rate of Covid-19 in the high risks countries. Chaos, Solitons & Fractals, volume.140, pages.110230, 2020, Elsevier.
[19] , Gupta, Vedika and Jain, Nikita and Katariya, Piyush and Kumar, Adarsh and Mohan, Senthilkumar and Ahmadian, Ali and Ferrara, Massimiliano. An emotion care model using multimodal textual analysis on COVID-19., Chaos, Solitons & Fractals, volume.144,pages.110708,2021, Elsevier.
[20] Zamir, Muhammad and Shah, Kamal and Nadeem, Fawad and Bajuri, Mohd Yazid and Ahmadian, Ali and Salahshour, Soheil and Ferrara, Massimiliano. Threshold conditions for global stability of disease free state of COVID-19. Results in Physics, volume.21, pages.103784, 2021, Elsevier.
[21] COVID, GitHub. database (2021) Retrieved from https://github. com/CSSEGISandData. COVID-19, year=19.
[22] Géron, Aurélien.
Hands-on machine learning with Scikit-Learn, Keras, and TensorFlow: Concepts, tools, and techniques to build intelligent systems.
2019, O’Reilly Media
[23] , Hochreiter, Sepp and Schmidhuber, Jürgen. Long short-term memory. Neural computation, volume.9, number.8, pages.1735–1780, 1997, MIT Press.
[24] Hassan, Syed Adeel and Sheikh, Fahad N and Jamal, Somia and Ezeh, Jude K and Akhtar, Ali. Coronavirus (COVID-19): a review of clinical features, diagnosis, and treatment. Cureus,volume.12,number.3,2020, Cureus Inc..
[25] , Singhal, Tanu. A review of coronavirus disease-2019 (COVID-19). The indian journal of pediatrics, volume.87,number.4,pages.281–286, 2020, Springer
[26] Wang, Wenzhao and Enilov, Martin. The global impact of COVID-19 on financial markets. Available at SSRN 3588021, 2020.
[27] Shah, Kamal and Khan, Zareen A and Ali, Amjad and Amin, Rohul and Khan, Hasib and Khan, Aziz. Haar wavelet collocation approach for the solution of fractional order COVID-19 model using Caputo derivative. Alexandria Engineering Journal,volume.59,number.5,pages.3221–3231, 2020, Elsevier.
[28] Khan, Arshad Alam and Ullah, Saif and Amin, Rohul. Control analysis of COVID-19 vaccine epidemic model: a case study, The European Physical Journal Plus, volume.137, number.1, pages.1–25, 2022, Springer.
[29] Khan, Arshad Alam and Amin, Rohul and Ullah, Saif and Sumelka, Wojciech and Altanji, Mohamed. Numerical simulation of a Caputo fractional epidemic model for the novel coronavirus with the impact of environmental transmission. Alexandria Engineering Journal,volume.61, number.7,5083–5095,2022,Elsevier.
[30] Negi, Alok and Kumar, Krishan and Chauhan, Prachi and Rajput, RS. Deep neural architecture for face mask detection on simulated masked face dataset against COVID-19 pandemic. 2021 International Conference on Computing, Communication, and Intelligent Systems (ICCCIS), pages.595–600,2021,IEEE.
[31] Negi, Alok and Chauhan, Prachi and Kumar, Krishan and Rajput, RS. Face mask detection classifier and model pruning with keras-surgeon. 2020 5th IEEE International Conference on Recent Advances and Innovations in Engineering (ICRAIE),pages=1–6,year=2020,IEEE
[32] Kumar, Krishan and Shrimankar, Deepti D. F-DES: Fast and deep event summarization. IEEE Transactions on Multimedia, volume.20,number.2,pages.323–334,2017,IEEE.
[33] Alok, Negi and Krishan, Kumar and Chauhan, Prachi. Deep learning-Based image classifier for malaria cell detection. Machine Learning for Healthcare Applications,pages.187–197,2021,Wiley Online Library.
[34] Ansari, Haroon and Vijayvergia, Aditya and Kumar, Krishan. DCR-HMM: Depression detection based on Content Rating using Hidden Markov Model. 2018 Conference on Information and Communication Technology (CICT), pages.1–6, year.2018, IEEE.
[35] Darbari, Anshuman and Kumar, Krishan and Darbari, Shubhankar and Patil, Prashant L. Requirement of artificial intelligence technology awareness for thoracic surgeons. The Cardiothoracic Surgeon, volume.29,number.1,pages.1–10,2021, SpringerOpen.
[36] Negi, Alok and Kumar, Krishan and Chauhan, Prachi Deep Neural Network-Based Multi-Class Image Classification for Plant Diseases Agricultural Informatics: Automation Using the IoT and Machine Learning,Pages.117–129,2021,Wiley Online Library.
[37] Negi, Alok and Kumar, Krishan Classification and detection of citrus diseases using deep learning Data science and its applications, pages.63–85,2021,Chapman and Hall/CRC
[38] Negi, Alok and Kumar, Krishan . Face Mask Detection in Real-Time Video Stream Using Deep Learning Computational Intelligence and Healthcare Informatics, pages.255–268,2021,Wiley Online Library.
[39] Negi, Alok and Kumar, Krishan and Chaudhari, Narendra S and Singh, Navjot and Chauhan, Prachi Predictive Analytics for Recognizing Human Activities Using Residual Network and Fine-Tuning. International Conference on Big Data Analyticspages.296–310, 2021, Springer.
[40] , Kumari, Shreya and Singh, Maheep and Kumar, Krishan Prediction of liver disease using grouping of machine learning classifiers International Conference on Deep Learning, Artificial Intelligence and Robotics, pages.339–349, 2019, Springer.
[41] Dabral, Ishani and Singh, Maheep and Kumar, Krishan.
Cancer detection using convolutional Neural network.
International Conference on Deep Learning, Artificial Intelligence and Robotics, pages.290–298,2019,Springer.
[42] Xu, Qingzhen A novel machine learning strategy based on two-dimensional numerical models in financial engineering Mathematical Problems in Engineering, volume.2013, year.2013, Hindawi.
[43] Xu, Qingzhen and Wang, Fengyun and Gong, Yongyi and Wang, Zhoutao and Zeng, Kun and Li, Qi and Luo, Xiaonan. A novel edge-oriented framework for saliency detection enhancement. Image and Vision Computing, volume.87, pages.1–12, year.2019, Elsevier
[44] Xu, Qingzhen and Huang, Guangyi and Yu, Mengjing and Guo, Yanliang. Fall prediction based on key points of human bones. Physica A: Statistical Mechanics and its Applications, volume.540, pages.123205,2020,Elsevier.
[45] Xu, Qingzhen and Wu, Jiayong and Chen, Qiang. A novel mobile personalized recommended method based on money flow model for stock exchange. Mathematical Problems in Engineering, volume.2014, 2014, Hindawi.
[46] Xu, Qingzhen and Wang, Zhoutao and Wang, Fengyun and Gong, Yongyi. Multi-feature fusion CNNs for Drosophila embryo of interest detection. Physica A: Statistical Mechanics and its Applications, volume.531,pages.121808, 2019,Elsevier.

Impact analysis of recovery cases due to COVID19 using LSTM deep learning model

Forecasting Economics and Financial Time Series: ARIMA vs. LSTM

Precipitation Nowcasting: Leveraging bidirectional LSTM and 1D CNN

DeepCOVIDNet: An Interpretable Deep Learning Model for Predictive Surveillance of COVID-19 Using Heterogeneous Features and their Interactions

Metamorphic Testing of a Deep Learning based Forecaster

Long-Short Ensemble Network for Bipolar Manic-Euthymic State Recognition Based on Wrist-worn Sensors

An Applied Deep Learning Approach for Estimating Soybean Relative Maturity from UAV Imagery to Aid Plant Breeding Decisions

Time Series Forecasting of New Cases and New Deaths Rate for COVID-19 using Deep Learning Methods

1 Introduction

2 Related Work

3 Proposed Methodology

3.1 Data Collection and Description

3.2 Feature Scaling

3.3 Architecture of Long Short Term Memory (LSTM)

4 Results and Discussion

5 Conclusions

Declarations

References

Impact analysis of recovery cases due to COVID19 using LSTM deep learning model

Related Research

Forecasting Economics and Financial Time Series: ARIMA vs. LSTM

Precipitation Nowcasting: Leveraging bidirectional LSTM and 1D CNN

DeepCOVIDNet: An Interpretable Deep Learning Model for Predictive Surveillance of COVID-19 Using Heterogeneous Features and their Interactions

Metamorphic Testing of a Deep Learning based Forecaster

Long-Short Ensemble Network for Bipolar Manic-Euthymic State Recognition Based on Wrist-worn Sensors

An Applied Deep Learning Approach for Estimating Soybean Relative Maturity from UAV Imagery to Aid Plant Breeding Decisions

Time Series Forecasting of New Cases and New Deaths Rate for COVID-19 using Deep Learning Methods

1 Introduction

2 Related Work

3 Proposed Methodology

3.1 Data Collection and Description

3.2 Feature Scaling

3.3 Architecture of Long Short Term Memory (LSTM)

4 Results and Discussion

5 Conclusions

Declarations

References