Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

Voting Regression Model for Covid-19 Time Series Data

Analysis

Saumya Gupta, Research Scholar, CSE, Dr. Chander Prabha, Associate Professor,
Chandigarh University, Mohali CSE, Chandigarh University, Mohali
(gupta98sept@gmail.com) (chander.e9251@cumail.in)

Abstract transparent reporting and multiple model assessments for the


prediction of future trends of this disease. With the ongoing
The cases of Coronavirus Disease 2019 are increased at a issues, it becomes challenging to predict the disease for
quick rate in mid-January and the virus is transmitted beyond analyzing the current situation and growth of COVID- 19
the borders of China. This disease is spread over 188 more efficiently so that the future of this ongoing pandemic
countries all around the world. The ARIMA model is an can be prepared. The suitable models are employed and
extensively utilized time series model that is constructed for accurate projections are generated for improving the public
economics applications. This model is adopted and becomes health decision-making; to illustrate, intervention-induced
famous due to its statistical properties. The covid-19 time changes in the transmission of disease.
series data analysis is a combination of three tasks such as
data acquisition, feature extraction, and classification. In this Various time series prediction models are present for
research work, voting regression is applied for the time-series analyzing the situation and trend of Coronavirus Disease
data analysis. The voting regression is the combination of the 2019. The time series prediction is a forecasting field whose
elastic net, linear regression, and lasso. The planned paradigm major intend is to analyze the earlier perceptions of an
is implemented in python and output is analyzed in terms of irregular variable for generating a paradigm that can capture
MSE and RMSE. The proposed model shows that MSE and the underlying association and its prototypes. Thereafter, the
RMSE values are low as compared to other models. subsequent values of that irregular variable are predicted
using this model. This concept is mainly utilized in a couple
Keywords of scenarios [2]. The first case has almost no information on
the fundamental information creating appropriation/ measure
Gradient Boosting, Linear regression, Xgboot regression, or the second case has not any explanatory model for
Voting regression, Covid-19 associating the predictive variable to other explanatory
variables in an adequate manner. Numerous efforts and
1. Introduction
research are conducted on constructing and improving the
Cases of severe respiratory illness have been noticed all over time series models over the previous years. The time series
the city of Wuhan in China in December 2019. This disease techniques having the influence of statistical and ML
is occurred due to a novel kind of coronavirus, SARS-Co V2 (machine learning) are deployed by several researchers for
that is known as COVID-19. The cases of Coronavirus estimating the percentage of active cases on the basis of the
Disease 2019 are increased at a quick rate in mid-January and complete populace for the ten nations with the most covid
the virus is transmitted beyond the borders of China. This positive patients. The ML and statistical techniques are
disease is spread over 188 countries all around the world. assisted in the past in the disease diagnosis. The modeling of
Instead of describing the effects of corona disease, various are leptospirosis and its association with rainfall and temperature
arise due to it such as the number of patients who are suffered can be taken as an example. Different time series-based
from novel coronavirus, the way to change the situation techniques are presented and analyzed such as Prophet,
change day by day and to predict the subsequent numbers of ARIMA, TBAT, HWAAS [3], N-Beats, and DeepAR. The
affected people with COVID-19 with the help of daily ARIMA model is an extensively utilized time series model
updated data on the trajectory of this disease. All of these that is constructed for economics applications. This model is
issues can be tackled with the prediction of possible futures adopted and becomes famous due to its statistical properties.
of this pandemic [1]. For this, the statistical or time series It employs Box–Jenkins’s methodology in the training phase
models can be implemented. But, these extensive systems and and this model is capable of deploying several exponential
tools may not provide higher accuracy while predicting the smoothing models. This model is based on the hypothesis that
disease, especially for medical studies. The evolvement of the time-series values are linearly correlated with the
Coronavirus Disease 2019 and the gathering of additional utilization of linear dependencies in observations for
data lead to construct and test the novel insights on the extracting the local patterns when the high-frequency noise is
novelty of the predictive models. But, there is a necessity for eliminated from the data. Three diverse input parameters are
exploited to describe this model in which 𝑎 denotes the lag the saturating point is not reached, are predicted using the
order or the number of lag observations included in the piece-wise model of steady development value which
model; 𝑑 is the number of times whose differentiation is done generates a productive and regularly valuable arrangement.
by the raw observations and𝑞 defines the size of the moving The seasonality is captured using the prophet that is
average window which assists in calculating the mean. In depending upon the Fourier series for creating a flexib le
case,𝑎 = 0, 𝑑 = 2, 𝑞 = 2, the forecast value 𝑌̂𝑡 which is model of periodic effects [7]. On the other hand, there is a
predicted with ARIMA is equal to: need for an already created index of historic and future break
𝑌̂𝑡 = 2𝑌𝑡−1 − 𝑌𝑡−2 − 𝜃1 𝑒𝑡−1 − 𝜃2 𝑒𝑡 −2 episodes for considering the holidays. Holiday effects are
considered independent. Thus, integrating them with the
In which,𝜃1 and 𝜃2 represent the moving average coefficients model is a challenging task. DeepAR is a predictive technique
and the forecasting error for the observation in time 𝑡 is planned on the basis of ARNN (autoregressive recurrent
defined with 𝑒𝑡 . This predictive technique is tested for the neural networks) for probabilistic forecasting. The suitable
selected data as the strong auto-correlation among data is probabilities are integrated with the nonlinear data
considered in it [4]. In addition, ARIMA can be deployed at transformation methods that are learned through DNN (deep
various kinds of datasets and provides optimal outcomes. The neural network) to carry out the prediction. The RNN
HWAAS is an extended version ofHES (Holt’s exponential structure based on LSTM is employed in this technique and
smoothing) that is a time series technique for univariate data. this technique is constructed based on DL (deep learning) for
This modification helps in adding the seasonal aspect to the time series data so that the issue of the probabilistic prediction
in-trend foresee to model knowledge with an orderly pattern can be resolved. N-Beats is a times series technique in which
or occasional part. The Holt-Winters additive model is a deep neural architecture is utilized and forward and backward
common technique that is able to tackle trends and weather remaining connections are involved with an in-depth heap of
changes. This model can be implemented as an option in layers that are fully linked [8]. In general, the execution of
contrast to the mainstream Box–Jenkins ARIMA group of this model is done as the conventional methods such as the
strategies. Nevertheless, experiential works have carried out seasonality-trend-level approach. There are two stacks of this
this technique isn't exact on normal as the more muddled system namely the trend stack and the seasonality stack.
Box–Jenkins technique is. The Holt–Winters’s additive Various blocks connected through the residual connections
model is efficient to be implemented on data having course are comprised in these stacks. This dual residual stacking in
and seasonal which stays stable with time and provides integration with the forecast principle provides the trend
bowed estimate to display the weather-based data variations component that is detached from the input window prior to its
[5]. The on-ground challenges faced during the deployment transmission into the seasonality stack. Thus, the decoupling
of this technique are the decision of introductory qualities, of partial forecasts of trend is done from the seasonality and
their affectability to uncommon occasions or anomalies, the it is presented as a distinct outcome using which an attractive
determination of smoothing metrics , and the standardization layer of significance is offered to the paradigm. Moreover,
of climatic index. The TBAT refers to an abbreviation for the this model can be trained quickly and provides interpretability
4 fundamental elements of the technique. The approach is and accuracy on various datasets.
depending upon the trigonometric functions exploit for
modeling the non-integer cyclical frequencies and Box-Co x 2. Literature Review
translations that are applied for transforming the atypical
dependent variables in a common shape and permit some Tingzhen Liu, et.al (2020) developed a transformation-based
forms of non-linearity. This technique is efficient for dealing autocorrelation sequence prediction algorithmic approach
with various issues related to time series. It is mathematically [9]. It constructed a time series of areas -based reference
successful for greatest probability assessors and its projected functions that had experienced the entire cycle of the
trigonometric formula is useful for decomposing the intricate pandemic condition. The ensembles of the Reference function
weather-based time series. This division is done to recognize were converted with other fields of incomplete observational
and extract the seasonal elements that are unseen in the time data as observables. The transformed function could
series plot. Facebook has introduced the Prophet a time series successfully predict series values hidden in other fields.
technique and designed for addressing the issues regarding Based on this, this work showed the way of using information
business time series [6]. The technique makes the delivered by other autocorrelation prediction-based
implementation of an easily decomposable time-series model exogenous variables to improve the output generated by the
in which trend, seasonality, and holidays are comprised as model. This work tested the new model using the COVID - 19
trend elements. This is a regression model having pandemic data set collected from Baidu. In the results, this
interpretable attributes which provide good performance with model showed high-quality fitting parameters. This model
their default values and assists the user in selecting the outperformed the basic LSTM pandemic predictive model.
components to handle their prediction issue and effortlessly
employ the necessary adjustments. The future trend is SinaArdabili, et.al (2020) performed a review with the intent
predicted using two systems namely a saturating growth to incorporate an ANN (artificial neural network)combined
model and a piece-wise linear model. In order to accomplish with the GWO (Gray Wolf Optimizer) approach for
the growth forecasting, a paradigm akin to the population predicting COVID-19 pandemic using a universal dataset
development paradigms in common environments is [10]. This work used time-series data collected to perform
deployed that contains the nonlinear growth to reach a training and testing processes, and verification. To evaluate
saturation point at a carrying capacity. The issues, at which the outcomes, this work employed MAPE and R values. The
presented approach yielded Mean Absolute Percentage Error other forms of data, such as mitigation measures, travel
around 6.23 for training, 13.15 for testing, and 11.4% for the effects, etc.
validation stages. The results showed that the devised model
could efficiently deal with the prediction work. Leonardo Sestrem de Oliveira, et.al (2021) implemented the
ANN model to predict the active number of Corona Virus
Raghavendra Kumar, et.al (2021) put forward a proposal of Disease 2019 cases and deaths in the next 7 days using the
using time series models for analyzing the outbreak of Corona time series of three countries [15]. The simulations could be
Virus Disease 2019 based on data of JHCRC (John Hopkins used to establish that the ANN model fruitfully predicted the
Coronavirus Resource Center) [11]. The Time Series model active number of cases along with the total no. of deaths due
predicted the active cases of corona disease in India with the to Covid-19. In general, an Artificial Neural Network with a
implementation of three time series models: ARIMA, MA, detailed test set had a 50% higher MSE as compared to an
and AR. Time series models could predict the current Artificial Neural Network with a randomized testing set. The
situation of the novel coronavirus pandemic. The new model LM training function obtained the least mean squared error
could help all types of organizations as a corresponding for overall cases.
technique to build and act upon decision-making strategies.
The impact of the pandemic on the whole nation could be Massimo A. Achterberg, et.al (2020) discussed that the
minimized using this model. This model provided accurate researchers belonged to diverse scientific fields focused on
prediction results which could probably be extended and predicting the spread of coronavirus disease 2019 [16]. An
validated utilizing other ML (machine learning) and DL (deep approach was recommended in which epidemic prediction
learning) models. techniques were comprised such as form-fitting techniques to
ML (machine-learning) models. A model known as NIPA
(Network Inference-based Prediction) was achieved by
Andi Sulasikin, et.al (2020) predicted the active cases of
integrating these techniques. This approach was utilized for
Covid-19 in the capital of Indonesia using two-time series
analyzing distinct sets of algorithms to predict the
models: ARIMA and Holt's exponential smoothing [12]. This coronavirus disease considering various modified versions of
work explored data and performed a comparative study of the presented model. It was observed that the original NIPA
time series models to ascertain the most competent models model provided superior performance to predict the spread of
for predicting the active cases of novel coronavirus 2019. COVID-19 in Hubei, China as compared to others. Moreover,
According to the results, the ARIMA model provided the the prediction based on the network outperformed the other
maximu m R-squared (R2 ), and the minimal MSE and RMSE algorithms.
for predicting the possible number of active cases of Corona
Virus Disease 2019 in Indonesia’s capital. The used system Md Masud Rana, et.al (2020) suggested a novel stochastic
performed remarkably well and generated appropriate system of coronavirus disease so that the communication
predictions that supported data-driven strategy in the context systems were prevented during the transmission and storage
of public health. of data [8]. The suggested system was constructed and
authenticated on the basis of observable and non-observable
states of corona disease including infection, confirmed or
Zehua Yu, et.al (2021) proposed a new model, namely IT-
unconfirmed cases, recovery, and deaths. A new secure
GCN to study the data of the Covid-19 outbreak [13]. In communication system based on IoT (internet of things) was
particular, the new model integrated ARIMA into GCN for introduced in order to send the local observation information
data modeling which originated on nodes in a graph that to the central control center. The coronavirus disease was
indicated the seriousness of the outbreak in many regions. forecasted by building and deploying an optimal signal
This work constructed the graph nodes with the vectors using processing algorithm. This approach emphasized mitigating
Auto-Regressive Integrated Moving Average the predictive error so that an efficient predictive system was
parameterization to discover the interaction topology in data. constructed. The outcomes demonstrated that the suggested
The outcomes showed that the presented model provided system was capable of predicting the states of coronavirus
better performance for predicting daily cas es of corona disease in a short period.
disease of short-term. The new model performed better than
the existing algorithm in the context of Mean Absolute Error, Chamara Sandeepa, et.al (2020) aimed to track the social
Root mean square error, and Mean Absolute Percentage interaction of users and predict the infection possibility on the
Error. basis of social interactions [9]. Initially, a social interaction
tracking system based on BLE (Bluetooth Low Energy) and
GPS was designed. Subsequently, an algorithm was
Radu Beche, et.al (2020) presented an auto-encoder implemented with the objective of predicting the possibility
prediction model to predict the active number of Corona of infected patients due to coronavirus disease 2019 in
Virus Disease 2019 cases in many European nations [14]. The accordance with the collected data. Finally, a mobile app and
output of this work could be used to assist officials to plan a web monitoring tool had employed to exploit a prototype of
preventive measures earlier and distribute resources the system. The simulations were conducted for the analysis
competently to minimize the effect of the new infection in of the behavior of the presented algorithm. The outcomes
their region. It was possible to apply the presented technique revealed that to slow down the impact of disease, self-
for foresting the outbreak of COVID-19 disease. It was isolation was essential.
noteworthy that the proposed model used merely
mathematical data without considering the context of any 3. Research Methodology
This research work is based on the covid-19 for time series LR also comes under the category of supervised learning
analysis. The time-series data analysis has various steps algorithms. This implies that the training of the model is done
which are described below: - on a set of labeled data. This model is utilized for forecasting
the labels on unlabeled data. Hierarchical partitioning of the
1. Dataset Collection and Pre-Processing: - The dataset is data is generated in which the diverse partitions at the leaf
collected from the kaggle. The dataset is of time series fro m level are related to the different classes with the help of DT.
the world cases. The target set of the dataset contains two A split criterion is carried out for producing hierarchica l
classes which are recovered cases, death cases. partitioning at every level. A condition may utilize a single
feature or multiple features in it. The first is called a
2. Feature Extraction and Regression: - The feature univariate split on the other hand second is named a
extraction is the second phase in which a relationship is multivariate split. The entire approach is utilized for
established between attribute and target set. The voting recursively splitting the training data so that the
regression model is applied for the time-series data analysis. discrimination is enlarged among the dissimilar classes over
The voting regression is the combination of the elastic net, diverse nodes. This bias is increased among the different
lasso, and linear regression. LASSO has expanded by classes in case of maximization of the level of skew among
presenting an elastic net regression model for which some of the different classes in a provided node. This skew is
its drawbacks are tackled in terms of selecting the variable. In computed using a measure named the Gini-index or entropy.
general, a grouping effect is influenced by this model for To illustrate, when 𝑃1 , … . . 𝑃𝑘 is the fraction of the records
which the predictors are co-related in this model. On the comes under the 𝑘 diverse classes in a node denoted by 𝑁, the
contrary, the strongest variable is kept in the Least Absolute Gini-index 𝐺 (𝑁 ) of the node 𝑁 is described as:
Shrinkage and Selection Operator (LASSO) for dividing
𝑘
these kinds of groups. In addition, the elastic net is proved
efficient at the time in case the number of predictors (p) is 𝐺 (𝑁 ) = 1 − ∑ 𝑃𝑖 2
greater within a dataset in comparison with the number of 𝑖=1

observations. However, LASSO has not potential to select


more than n predictors. Thus, the elastic net model is proved The value of 𝐺 (𝑁 ) lies between 0 and 1 − 1/𝑘. As the value
more efficient. Suppose the standardized regressors and of 𝐺 (𝑁 ) becomes small, the skew becomes greater. If the
response, the below-given issue can be solved through the classes are balanced in even order, the value is 1 − 1/𝑘.A
elastic net as: Voting Regression is a technique whose training is executed
on an ensemble of a variety of models. This algorithm assists
𝑁
2 in predicting an output based on the highest probability of the
selected class as the output. This regression technique is
𝛽̂ = arg 𝑚𝑖𝑛 {∑ (𝑦𝑖 − ∑ 𝛽𝑗 𝑥 𝑖𝑗 ) + 𝜆𝑃𝛼 (𝛽) }
capable of aggregating the discoveries of every regression
2𝑁 𝑗
process that is gone through the voting model. The outcomes
attained from various classification methods are integrated
1
= arg min { ‖𝑦 − 𝑋𝛽‖ 22 + 𝜆𝑃𝛼 (𝛽) } into this algorithm. The voting regression algorithm has three
2𝑛
categories. All the classification schemes are agreed with the
final decision in unanimous voting. The majority voting
Pα (β) = α ‖β‖22 + (1 − α)‖β‖ 1 makes the final decision on the basis of more than 50% votes.
𝑝
The plurality voting technique is very simple that facilitated
= ∑ 𝛼𝛽𝑗2 + (1 − 𝛼) |𝛽𝑗 |
every each voter for selecting only one alternative. The
𝑗=1
candidate who attains the highest number of votes is
considered the winner. In the case of m voters who are
In which, Pα(β) denotes the elastic net penalty. Pα (β) is selected through simple majority voting and the probability ,
utilized to represent the integration of the ridge regression 𝑝 is considered for every voter in order to make the accurate
penalty, for α = 1 with the LASSO penalty, for α = 0. Such a decision, the probability of the whole jury to take the right
form of penalty is often proved helpful when p > n or in decision is defined as:
situations that includes a number of correlated predictor
variables. The LR as part of the family of regression 𝑚
𝑚!
algorithms emphasizes discovering relationships and 𝑝𝑚 = ∑ ( ) ∙ 𝑝 𝑖 ∙ (1 − 𝑝) 𝑚−1
dependence among between variables. A modeling (𝑚 − 𝑖) ! ∙ 𝑖!
𝑖= ⌈𝑚 ⁄2 ⌉
relationship is demonstrated amid a continuous scalar
dependent variable represented by y and one or more Therefore, in case 𝑝 > 5.0, then 𝑝𝑚 > 𝑝.This implies that a
explanatory variables in which other variables are contained higher probability to make an accurate decision through an
revealed with 𝑋 through a linear function. The major intend ensemble voting regression in comparison with any
of the regression analysis is that a continuous target variable individual voter.
has to be forecasted while a label from a finite set is utilized
to predict the other region known as classification. The model
employed to perform the Multiple Regression in which input 4. Result and Discussion
variables are expressed as:
This research work is based on the covid-19 time series data
𝑦 = 𝛽0 + 𝛽1 𝑥 1 + 𝛽2 𝑥 2 +. . . +𝑒 analysis. The dataset is collected from the kaggle for the time
series analysis. The dataset contains a total of 12 columns in
which the first 11 are the attributes and the last one is the
target set. The attributes are identification, country, Province,
Country region, population, weight, date, target. The target
value is the last attribute that defines the target set. The values
attribute like weight, data, identification is dropped because
these are used to keep records in the hospital and not used for
the prediction analysis.

The times series dataset is analyzed using various regression


models like gradient boosting, linear regression, XGboost,
and the proposed model. The gradient boosting, linear
regression, and XGboost are the existing models which are
used for the time series data analysis. The proposed model is
Figure 1: MSE analysis
the voting regression model which is the combination of
linear regression, lasso, and elastic net.
As shown in figure 1, the MSE values of the gradient
boosting, linear regression, xgboost, and proposed technique
The performance analysis parameters are explained below:-
is compared for the performance analysis. It is analyzed that
the MSE value of the proposed method is low in comparison
1. Root Mean Square Error:- It is the standard deviation of
with existing methods.
the prediction errors. These errors define the distance of data
points from the regression line. This value computes the
reason for the transmission of these residuals. Moreover,
RMSE described the way to find out the data around the line
of best fit.

2. Mean Squared Error:- This value defines the close


relation of the regression line to a set of points. For this, the
distance amid the points and the regression line is computed
and its square is found. Any negative sign can be eliminated
with the squaring. MSE provides more weight to larger
differences.

Table 1: Performance Analysis


Figure 2: RMSE value Analysis
Para Gradie Linear Xgboost Propos
meter nt Regressi Regressi ed
Boosti on on As shown in figure 2, the RMSE value of gradient boosting,
ng linear regression, xgbosst, and the proposed algorithm is
MSE 0.010 0.018 0.087 0.003 compared for the performance analysis. The proposed
algorithm has the least RMSE value for the covid-19
prediction.
RMS 1.32 0.18 0.29 0.13
E Conclusion

Covid disease is spread over 188 countries all around the


As shown in Table 1, the various models like gradient world. Instead of describing the effects of corona disease,
boosting, linear regression, xgboost regression, and proposed various are arise due to it such as the number of patients who
models are compared for the analysis of the timer series are suffered from COVID-19, the way to change the situation
dataset. The performance of the models is compared in terms change day by day, and for predicting the future numbers of
of MSE and RMSE. The proposed model gives maximu m affected people with corona disease with the help of daily
performance MSE and RMSE is 0.06 and 0.13 respectively updated data on the trajectory of this disease. The Covid
due to hybridization. The proposed model uses the voting disease prediction has diverse stages in which pre-processing
regression which is the combination of linear regression, is done, attributes are extracted and classification is
lasso, and elastic net for the analysis of the time-series dataset performed. In this research, a voting regression model is
of COVID-19.
proposed which is the combination of linear regression,
elastic net, and decision tree. The proposed model is
implemented in python and results will be analyzed in terms
of MSE and RMSE. It is analyzed that the MSE and RMSE
value of the proposed model is low as compared to existing
models.
Reference Global Prediction Using Hybrid Artificial Intelligence
Method of ANN Trained with Grey Wolf Optimizer”, 2020,
[1] Saud Shaikh, Jaini Gala, Aishita Jain, Sunny Advani, IEEE 3rd International Conference and Workshop in Óbuda
Sagar Jaidhara, Mani Roja Edinburgh, “Analysis and on Electrical and Power Engineering (CANDO-EPE)
Prediction of COVID-19 using Regression Models and Time
Series Forecasting”, 2021, 11th International Conference on [11] Raghavendra Kumar, Anjali Jain, Arun Kumar Tripathi,
Cloud Computing, Data Science & Engineering (Confluence) Shaifali Tyagi, “COVID-19 Outbreak: An Epidemic Analysis
using Time Series Prediction Model”, 2021, 11th
[2] Arif Kurniawan, Fachrul Kurniawan, “Time Series International Conference on Cloud Computing, Data Science
Forecasting for the Spread of Covid-19 in Indonesia Using & Engineering (Confluence)
Curve Fitting”, 2021, 3rd East Indonesia Conference on
Computer and Information Technology (EIConCIT) [12] Andi Sulasikin, YudhistiraNugraha, Juan Kanggrawan,
Alex L. Suherman, “Forecasting for a data-driven policy
[3] Mogens Graf Plessen, “Integrated Time Series using time series methods in handling COVID-19 pandemic
Summarization and Prediction Algorithm and its Application in Jakarta”, 2020, IEEE International Smart Cities
to COVID-19 Data Mining”, 2020, IEEE International Conference (ISC2)
Conference on Big Data (Big Data)
[13] Zehua Yu, Xianwei Zheng, Zhulun Yang, Bowen Lu,
[4] Suraj Bodapati, Harika Bandarupally, M Trupthi, Xutao Li, Maxian Fu, “Interaction-Temporal GCN: A Hybrid
“COVID-19 Time Series Forecasting of Daily Cases, Deaths Deep Framework For Covid-19 Pandemic Analysis”, 20021,
Caused and Recovered Cases using Long Short Term IEEE Open Journal of Engineering in Medicine and Biology
Memory Networks”, 2020, IEEE 5th International
Conference on Computing Communication and Automation [14] Radu Beche, Romina Baila, Anca Marginean, “COVID -
(ICCCA) 19 spread forecast using recurrent auto-encoders”, 2020,
IEEE 16th International Conference on Intelligent Computer
[5] Naresh Kumar, Seba Susan, “COVID-19 Pandemic Communication and Processing (ICCP)
Prediction using Time Series Forecasting Models”, 2020,
11th International Conference on Computing, [15] Leonardo Sestrem de Oliveira, Sarah Beatriz
Communication and Networking Technologies (ICCCNT) Gruetzmacher, João Paulo Teixeira, “COVID-19 Time Series
Prediction”, 2021, Procedia Computer Science
[6] Mohsen Mousavi, Rohit Salgotra, Damien Holloway ,
Amir H. Gandomi, “COVID-19 Time Series Forecast Using [16] Massimo A. Achterberg, Bastian Prasse, Piet Van
Transmission Rate and Meteorological Parameters as Mieghem, “Comparing the accuracy of several network-
Features”, 2020, IEEE Computational Intelligence Magazine, based COVID-19 prediction algorithms”, 2020, International
Volume: 15, Issue: 4 Journal of Forecasting Available online

[7] VenkatbharatPoleneni, Jahnavi K Rao, Syed [17] Md Masud Rana, Ahmed Abdelhadi, Md Riaz Uddin
AfshanaHidayathulla, “COVID-19 Prediction using ARIMA Ahmed, Ahad Ali, “Secure IoT Communication Systems for
Model”, 2021, 11th International Conference on Cloud Prediction of COVID-19 Outbreak: An Optimal Signal
Processing Algorithm”, 2020, Third International Conference
Computing, Data Science & Engineering (Confluence)
on Smart Systems and Inventive Technology (ICSSIT)

[8] Abdul Jalil Niazai, Abdullah Zahirzada, Mohammad [18] Chamara Sandeepa, CharukaMoremada,
Akbar Shahpoor, Abdul Rahman Safi, “Time Series NadeekaDissanayaka, Tharindu Gamage, Madusanka
Forecasting of Registered, Recovered, and Death Cases of Liyanage, “Social Interaction Tracking and Patient Prediction
COVID-19 for the Next Sixty Days in Afghanistan”, 2020, System for Potential COVID-19 Patients”, 2020, IEEE 3rd
IEEE International Conference on Advent Trends in 5G World Forum (5GWF)
Multidisciplinary Research and

[9] Tingzhen Liu, Tong Zhou, Jin Gao, Wei Li, Yimin Ma,
“Autocorrelation Sequence Prediction Model Based On
Reference Function Transformation: Taking Epidemic
Prediction As An Example”, 2020, Chinese Automation
Congress (CAC)

[10] SinaArdabili, Amir Mosavi, Shahab S. Band, Annamaria


R. Varkonyi-Koczy, “Coronavirus Disease (COVID-1 9)

You might also like