Springer Nature 2021 LATEX template

Comparing short-term univariate and

multivariate time series forecasting models in
infectious disease outbreak.
† †
, and
*,,,,,,.
,,,,,,.

*Corresponding author(s). E-mail(s): ;

† These authors contributed equally to this work.

Abstract
Predicting infectious disease outbreak impacts on population,
health care resources and economics has received a special
academic focus during coronavirus (COVID-19) pandemic.
Focus on human disease outbreak prediction techniques in current lit-
erature, Marques et al. [1] states that there are four main methods
to address forecasting problem: Compartmental Models, Classic Statis-
tical Models, Space State Models and Machine Learning Models. We
adopt their framework to compare our research with previous works.
Besides being divided by methods, forecasting problems can also be
divided by the number of variables that are taken into account to
make predictions. Considering this number of variables, forecasting
problems can be classified as univariate, causal and multivariate models.
Multivariate approaches have been applied in less than 10% of
research found. This research is the first attempt to evaluate,
over real time-series data of 3 different countries with univari-
ate and multivariate methods to provide a short-term prediction.
In literature we found no research with that scope and aim.

Keywords: Outbreak Disease, COVID-19, Forecasting, Univariate approach,

Multivariate approach

1
 Springer Nature 2021 LATEX template

2 Comparing short-term univariate and multivariate time series forecasting ...

1 Introduction
Communicable (or infectious) diseases can rapidly spread because it caused
by breathing in an airborne virus, bitten by insects, sexual intercourse, skin
contact by patient who is already suffer with that disease Kaur et al. [2]
When the disease spread runs out of control and infect a community or
region with specified health behavior, or other health-related events clearly
more than normal expectancy it is called epidemic [3]. The term pandemic
is commonly taken to refer to a widespread epidemic of contagious disease
throughout the whole of a country or one or more continents at the same time
[4].
Although personal measures should be taken to avoid the infection and
therefore their spread like do not share personal things, clean hands properly,
always take good and safe food, get vaccinated or cover month when sneezing
or coughing always cover your mouth [2], health systems and governments of
all countries must be able to develop and improve non pharmacological mea-
sures like animal source containment, early detection and diagnosis, rigorous
infection control, timely case report and rapid information dissemination, quar-
antines, mask obligation, lockdown and pharmacological measures like vaccine
development [5].
Over the last few decades, mathematical models applied over infectious
diseases growth have been helpful to gain insights into the transmission dynam-
ics [6] allowing to forecast new cases and deaths as well as evaluate the
interventions’ impact [7].
Besides, still showing numerous limitations and pitfalls, often driven by
data scarcity and delay, Smirnova and Chowell [8] state that integration of
mathematical models’ prediction results with public health practice has the
potential to increase the timeliness and quality of health care unit responses.
In this context, during COVID-19 pandemic, Marques et al. [1] applied four
univariate forecasting approaches using real COVID-19 data of 5 countries.
These approaches are Classical Statistical Models, Compartmental models,
State-Space Model and Machine Learning models and will be presented in
section 2.
After evaluating and comparing sixty six previous works (see Table A1),
we conclude that Less than 10 % of previous research applied multivariate
techniques and none of them were used more than one country or region. Thus,
this research contributes to forecasting methods application over human
infections diseases outbreaks by being the first attempt to evaluate, over real
time-series data:
• For three different countries (Brazil, Italy, and United States);
• Using six univariate and two multivariate methods;
• Providing a short-term prediction of 28 days ahead which is two or four
times longer than similar previous research;
 Springer Nature 2021 LATEX template

Comparing short-term univariate and multivariate time series forecasting ... 3

In section 3 we present the all time-series evaluated in this research and

their features, how we choose the data range for all time-series and how we
split these data in data training and data test.
In section 4 we briefly explain which forecasting methods applied in all
time-series and specify how the results are obtained and compared. After chose
the best model for each time-series we make a short-term prediction of 28 days.
Finally, in section 5 we present research’s conclusions, address limitations
and make proposals for further research.

2 Theoretical Background
Epidemics or pandemics disease outbreak have been devastating populations
worldwide all over the years [9],[10] and [2]. From Athens epidemics (“Plague
of Athens”) in 430-427 B.C (see Hays [9] for more details) to Corona Virus
(SARS-CoV 2) also known as COVID-19 on going pandemic, the civilizations
have lived with epidemics or pandemics caused mainly by virus and bacteria.
Kaur et al. [2] summarized the most relevant disease outbreaks in human
history like Blackdeath (black plague), cholera, malaria and influenzas virus
(Spanish, Hong Kong and Russian Flu). In addition, Hays [9], [10] and Yamey
et al. [11] points out many others like the smallpox, Blackdeath (black plague),
cholera, influenza, HIV/AIDS, measles, dengue, Ebola, Zika virus.
Table 1 summarizes in a non-exhaustive list of worldwide human outbreaks
diseases (epidemics or pandemics) by year, impact in number of deaths and
where each one occurred.
Besides the number of human deaths caused by epidemics and pandemics,
Kaur et al. [2] state that and it will not disappear in future if we do not
find efficient ways to stop before spreading any disease to other population or
countries.
To explain, evaluate and estimate further values (forecast) the behavior of
some variable like outbreak disease cases, deaths, or transmission rate all over
the time many authors use time-series approach [12] [13] [14] [15].
A time-series is a set of data points arranged in time and its analysis intends
to reveal reliable and meaningful statistics [1] that can be used to evaluate
some patterns and forecast future values [16]. It drew the attention of the
scientific community when Yule introduced a general approach for time-series
analysis in 1927 [17].
In the same year, one deterministic compartimental model widely applied
in epidemiology science was proposed by Kermack and McKendrick [18], the
Susceptible–Infectious–Removed (SIR) model.
Almost three decades later (1950s) classical time series statistical models
started to appear (Holt [19]; Brown [20]; Winters [21]; Box and Jenkins [22]) as
well as Machine Learning (Samuel [23]) and Space-State Model (Kalman [24]).
Bring forecasting methods to human infectious disease outbreak con-
text, Chretien et al. [25] proposed a framework to classify research as
follows: Population-based forecasting studies (seasonal or pandemic), forecast
 Springer Nature 2021 LATEX template

4 Comparing short-term univariate and multivariate time series forecasting ...

Table 1 Worldwide epidemics or pandemic. Source: The authors adapted from Yamey et
al. [11], Yang et al. [5], Kaur et al. [2].

Outbreak disease Year Impact Countries

75,000-100,000 deaths
Athens Plague 430-427 B. C Greece
(25% of population)
Asia Minor, Egypt,
Antonine Plague 165-180 5 million deaths
Greece, Italy
Justinian Plague 541-542 35 million deaths Eastern roman empire
50 million deaths
(60% of Europe
Black Plague 1346-1353 population). Europe, Africa and Asia
75-200 million
deaths worldwide
Asia, Africa, France,
Fifth Cholera 1881-1896 981,899 deaths Germany, Russia,
and South America
Russia, Canada
Russian Flu 1889-1890 1 million deaths
and Greenland
Modern Plague 1894-1903 10 million deaths India and China mostly
North Africa, Middle East,
Sixth Cholera 1899-1923 1.5 million deaths India, Eastern Europe,
Russia and America
Australia, Canada,
Spanish Flu 1918-1919 20-50 million deaths UK, US, France
and Spain
China, Singapore,
Asian Flu 1956-1958 2 million deaths
US and Hong Kong
Hong Kong and spread
Hong Kong Flu 1968-1969 1 million deaths Singapore, Vietnam, US
India, Australia and Europe
HIV/AIDS 1980-present 39 million deaths Worldwide
Severe Acute
8422 people infected in
Respiratory China, Vietnam, Singapore,
2002-2003 32 countries and 919
Syndrome Canada and others
(11%) died
(SARS CoV)
Mexico, US, Africa
Swine Flu 2009-2010 150,000-750,000 deaths
and Southeast Asia
Middle East
Respiratory
2012 858 Saudi Arabia
Corona virus
(MERS CoV)
Ebola 2013 10,600 deaths Guinea, Liberia, Sierra Leone
Zika Virus 2015-2016 20 deaths Latin America and Caribbean
Coronavirus
SARS CoV 2 2019-present 4.94 million deaths Worldwide
(COVID-19)

type (temporal or Spatial-temporal) and forecasting method (mechanistic,

Statistical).
According to the same author, The forecasting method were divided in
compartmental model, regression tree, generalized linear model, agent-based
model, survival analysis, Bayesian network and time series model.
 Springer Nature 2021 LATEX template

Comparing short-term univariate and multivariate time series forecasting ... 5

Focusing on forecasting method, during ongoing COVID-19 pandemic,

Marques et al. [1] presented four different univariate approaches for epidemio-
logical time-series prediction, which will be able to provide decision support for
Governments and Healthcare decision-makers. They worked with five countries
real data: China, United States, Brazil, Italy, and Singapore.
In this research we adopt the framework proposed by Marques et al. [1] that
divided epidemiological time-series prediction in: Classical Statistical mod-
els (section 2.1), Compartmental models (section 2.2), State-Space models
(section 2.3) and Machine learning models (section 2.4).

2.1 Classical Statistical Models (CSM)

In this section we present CSM methods found in literature that are divided in:
• Exponential smoothing (ES) or their generalization Error, Trend and
Seasonal (ETS);
• Autoregressive Integrated Moving Average (ARIMA);
• Vector Autoregressive (VAR);
• Vector Error Correction (VEC);
• Vector Autoregressive Moving Average (VARMA).
ES was proposed in the late 1950s (Holt, 1957; Brown, 1959; Winters,
1960), and has motivated some of the most successful forecasting methods.
ARIMA was introduced by Box and Jenkins [22] in the 1970 and takes into
consideration changing disturbances in time and tendencies.
Hyndman and Athanasopoulos [16] state that ES, or their generalization
ETS, and ARIMA models are the two most widely used approaches to time
series forecasting and provide complementary approaches to the problem.
While ES models are based on a description of the trend and seasonality in
the data, ARIMA models aim to describe the autocorrelations in the data.
In econometric research focusing on short term prediction multivariate
approach widely applied are VAR,VEC and VARMA.
Kiang et al. [26], Arunkumar et al. [13], Talkhi et al. [27], Katri [14], Khan
et al. [28], Bomfim et al. [29], Liang et al. [30], Ramos et al. [31], Zhang et al.
[32], Wang et al. [33], Li et al. [34], Choi et al. [35], Chakrab et al. [36], Chu-
machenko et al. [37], Haddaw et al. [38], Wu et al. [39], Wu et al. [40] Zhao et
al. [41], Jerónimo-Martinez et al. [42], Ray et al. [43], Anggraeni and Aristiani
[44], Ke et al. [45], Li et al. [46], Johansson et al. [47], Pradhan et al [48], Wu
et al. [49] Mekparyup and Saithanu [50], Kane et al. [51], Feng et al. [52], Soe-
biyanto et al. [53], Shen et al. [54], Medina et al. [55], Burkrom et al. [56] and
Nobre et al. [57] research applied these models to several human disease out-
breaks like COVID-19, Ebola, Zika Virus, Dengue Hemorraric Fever (DHF),
Scarlet Fever (SF), Tuberculosis, Malaria, leprosy, haemorragic fever with
renal syndrome (HFRS), hand, foot, and mouth disease (HFMD), HIV/AIDS,
tuberculosis, malaria, influenza-like illness (ILI) and others Acute Respiratory
infection (ARI).
 Springer Nature 2021 LATEX template

6 Comparing short-term univariate and multivariate time series forecasting ...

2.2 Compartmental Models (CM)

In this section we present CM methods found in literature that are divided in:
• Susceptible– Infectious–Removed (SIR);
• Susceptible-exposed-infectious-removed (SEIR);
• Susceptible-infectious-susceptible (SIS);
• Cellular Automation (CA);
• Growth Models (GM).
One deterministic model widely considered in epidemiology is the SIR
model, which is based on the classification of the individuals into three stages
of infection and was introduced almost one hundred years ago by Kermack
and McKendrick [18].
All over the years SIR model was improved, and other stages were added
[58], for instance: SEIR with or without intervention and SIS among others.
Considering single variables, GM like Richards (GMR), Gompertz (GMG),
Logistic (GML) and Cellular Automation (CA) are widely used [59] to describe
and predict infectious diseases spread cases and deaths.
Chen et al. [12], Katris [14], Paul et al. [60], Benı́tez et al. [15], Wang et al.
[61], Smirnova et al. [62], Eilterson et al. [63], Suparit et al. [64], Basile et al.
[65], Li et al. [66], Valeri et al [67], Yang et al. [68], Wang et al. [69], Towers
and Chowell [70], Aguiar et al [71], Gerardi and Monteiro [59], Laneri et al.
[72], Santos et al. [73], Finkenstädt et al. [74] and Gamerman and Migon [75]
research applied these models to several human disease outbreaks like COVID-
19, Measles, ILI, dengue, DHF, Skin and Soft Tissue Infections (SSTIS).

2.3 State-Space Models (SSM)

In this section we present SSM methods found in literature that are divided in:
• Hidden Markov Model (HMM);
• Monte Carlo Markov Chain (MCMC);
• Kalman filter (KF);
• Exponential smoothing state space model with Trigonometric, Box-Cox
transformation, ARMA errors, Trend and Seasonal components (TBATS).
A SSM, also known in the technical literature as HMM, can be defined
as a class of probabilistic models that describes the dependence between a
latent state variable and an observed measurement [76]. The term “state space”
originated in control engineering subject [24]. HMM can also be combined with
simulation approach like Monte Carlo. It is called, according to Wang et al.
[69] MCMC.
SSM is a general framework for ES, ARMA and Trend and Seasonal com-
ponent where TBATS, according to Talkhi et al. [27], is widely applied to
univariate time series.
The KF is a state-space model provides estimates of the unknown variables
given the measurements observed over time using only the previous estimate
 Springer Nature 2021 LATEX template

Comparing short-term univariate and multivariate time series forecasting ... 7

for calculation which reduces the need for saving the whole data from previous
iterations [77].
Talkhi et al. [27], Han et al. [78], Benı́tez et al. [15], Eilterson et al. [63],
Yang et al. [68], Wang et al. [69], Nunes et al [79], Mode et al. [80] research
applied these models to several human disease outbreaks like COVID-19,
Ebola, Zika Virus, ILI, SSTIS, HIV/AIDS.

2.4 Machine learning Models (MLM)

In this section we present MLM methods found in literature that are divided
in:
• Multi-layer perceptron (MLP);
• Artificial recurrent neural network (RNN);
• Long short-term memory (LSTM);
• Convolutional Neural Network (CNN);
• Feed-forward neural networks with a single hidden layer and lagged inputs
(NNETAR) that is also divided in Neural Network autoregressive (NNAR)
and Nonlinear Auto-Regressive Neural Network (NARNN);
• Extreme Learning Machine algorithm (ELM);
• Automated Machine Learning (AutoML);
• Ensemble empirical mode decomposition (EEMD);
• Cross-Location Attention Graph Neural Network (CLAGNN);
• Support Vector Machine (SVM);
• Bayesian Model Averaging (BMA);
• Kernel Conditional Density Estimation (KCDE);
• Kernel Ridge Regression Gausian Process Network (KRRGPN);
• Neural fuzzy inference system (NFIS);
• Random forest (RF);
• Generalized Regression Neural Network (GRNN);
• Genetic Algorithm (GA);
• Wavelet Neural Network (WNN).
First defined by computer scientists at the Dartmouth Conferences in 1956,
Artificial Intelligence (AI) field draws upon computer science, mathematics,
psychology, linguistics, neuroscience and many others [81].
Evolved from the study of pattern recognition and computational learning
theory in AI, Machine Learning (ML) appears in 1959 [23] to explore the study
and build algorithms that can learn from and make predictions on data [82].
Deep Learning (DL) and Neuro Networks (NN) are subfields of ML (for more
details see Alzubi et al [83] and Ongsulee [81]).
Some ML algorithms widely applied to forecast are MLP, RNN like
LSTM with or without false nearest neighbors (FNN), CNN, NNETAR,
ELM, AutoML, EEMD, CLAGNN, SVM, BMA, NARNN, NNAR, KCDE,
KRRGPN, NFIS, RF, GRNN, GA and WNN.
All algorithms mentioned above were applied in Arunkumar et al. [13],
Talkhi et al. [27], Katris [14], Han et al. [78], Ribeiro et al. [84], Deng et al.
 Springer Nature 2021 LATEX template

8 Comparing short-term univariate and multivariate time series forecasting ...

[85], Wang et al. [61], Bomfim et al. [29], Liang et al. [30], Zhang et al. [32],
Wang et al. [33], Choi et al. [35], Chakraborty et al. [36], Stolerman et al. [86],
Wu et al. [40], Ray et al. [43], Caicedo-Torres et al. [87], Nguyen et al. [88],
Chau and Ngoc [89], Wu et al. [49], Kane et al. [51], Gerardi and Monteiro
[59] and Peng et al. [90] research.

2.5 Research Synthesis

Table A1 summarize forecast research applied in human infectious disease
outbreak (pandemic or epidemic) context considering the “general methods”
pointed out by Marques et al. [1] the results. Only research that in fact make
predictions were considered.
Columns 3 to 7 address the approach used in each research mentioned in
last section. In columns 8 to 9 is presented the range of time windows used as
well as the prediction range. Time windows found on previous research were
day (d), week (w), month (m) or year (y). The prediction range are expressed
in time windows mentioned, but some models proposed to forecast the whole
pandemic period (wpp).
The variable measured/evaluated and forecasted in each research (column
10) on those time windows are number of patient cases (ca), deaths (de) and
recovered (re), admitted and discharged from hospital or intensive care unit
(adhosp and dishosp) and transmission rate (rt). Excluding rt, all measures
mentioned can be counted in two different ways: by time window or cumulative.
For example, daily cases (dca), monthly deaths (mde), yearly patients admitted
in hospital (yadhosp), cumulative cases (Cca), cumulative patients discharged
from hospital (Cdishosp).
Columns 11 to 13 presents in which countries each research applied the
methods specified in columns 3 to 7, the type of forecast approach divided
in Univariate (Uni), Causal or Multivariate (Mul) and the disease outbreak
studied.
After comparing sixty six works in Table A1 we can conclude that:
• Approach: only seven (10.61%) apply multivariate methods to make pre-
dictions. Twenty three (34,84%) apply causal methods to make predictions
combining number of specific queries all over the time on web search engines
(Google, Baidu index), climate variables (temperature, air pollution, rain-
fall) or another seasonality infectious diseases. thirty six (54,55%) apply
only univariate methods to make predictions. The number of publications
by approach is presented in Figure 1.
• Disease outbreaks: fourteen research forecasted Dengue, Chikungunya or
DHF, fourteen ILI, nine COVID-19, six HFRS, five Malaria, four Measles
three Ebola and three HIV/AIDS.
• Countries: fifteen research were applied in disease outbreak in China, nine in
Brazil, nine in US, four in Thailand and three in Japan. Only four research
worked with African countries.
 Springer Nature 2021 LATEX template

Comparing short-term univariate and multivariate time series forecasting ... 9

• Data range and prediction: only four research proposed to predict the whole
pandemic period (wpp). Considering different time windows, twenty six
research forecasted less or equal to six steps ahead.
• Time window: twenty eight research worked with monthly cases and twenty
four with weekly cases.
• Variables: number of patient cases was studied in sixty three research
(95,45%) while deaths, recovered and hospital admission or discharge or
transmission rate are not much explored (deaths appears in second place
with only five research).
• Epidemiological time-series prediction: thirty four research applied CSM,
twenty three applied MLM, twenty applied CM and only eight applied SSM.
We found no research in which all approaches were applied. The number of
publications by type of epidemiological time-series prediction is presented
in Figure 2. Only two research used three univariate approaches (Talkhi et
al., 2021 and Katris, 2021) in a single country (Iran and Greece respec-
tively), sixteen research used two approaches, forty five research used only
one approach and three reseach used approaches not mentioned by Marques
et al. (2021).
• On twenty CM models only five worked with growth models and basically
applied three models: Richards (GMR), Gompertz (GMG), Logistic (GML).
But we point out that there is other fourteen GM models (Fekedulegn et
al. [91], Kaps et al. [92], Tsoularis and Wallace [93] and Khamis et al. [94])
that were not explored in human disease outbreak context.
6

Uni
Causal
5

Mul
Number of publications

4
3
2
1
0

1990 1995 2000 2005 2010 2015 2020

years

Fig. 1 Research publication by approach by year. Source: authors.

 Springer Nature 2021 LATEX template

10 Comparing short-term univariate and multivariate time series forecasting ...

CSM
CM
6

SSM
Number of publications

MLM
4
3
2
1
0

1990 1995 2000 2005 2010 2015 2020

years

Fig. 2 Research publication by model by year. Source: authors.

Although the current review shows benefits of using CM models including

to provide medium- and long-term predictions and mostly uses Suscepti-
ble–Infectious–Removed models or their variations, many assumptions over
them must be made [62] before obtaining all parameters, results of all stages
and then a prediction of a whole pandemic period.
In addition, the real time data COVID-19 showed us that new stages
are necessary to be considered like immunity period and rate of reinfection,
vaccination, period of strong non-pharmacological measures (quarantine and
lockdown), among others.
The current research is the first attempt to evaluate over real data of three
different countries (Brazil, Italy, and US) using three univariate approaches
(CSM, SSM and MLM) proposed by Marques et al. (2021). We apply the same
univariate methods proposed in Talkhi et al. (2021) and add KF.
Finally, we apply two multivariate approaches and compare their results
with previous mentioned univariate methods to find which approach can better
fit real data and give us a reliable short-term prediction to each region/country.

3 Data sets Selection and Problem Statement

In this research we work with real COVID-19 data of Rio de Janeiro (RJ) [95]
city health regions in Brazil, Italy (IT) regions [96] and US states [97]. All
time-series used are presented in figures below.
We select these data sets because Brazil, Italy and US population were
highly affected by COVID-19 pandemic and adopted different rules to fight
against COVID-19 dissemination.
 Springer Nature 2021 LATEX template

Comparing short-term univariate and multivariate time series forecasting ... 11

After first COVID-19 wave starts to spread, Italian government establish

hard common measures for all country regions. Could we expect that the
number of daily new cases from one region helps to explain and predict this
number in other region or only the past data of the same region is better
enough?
In other words, given common government rules which approach best fits
and predict daily number of cases: univariate or multivariate methods?
Regional divisions of each Italian time-series used in this research are
quickly presented in Table 2. For more details see supplementary material.
In US, each State has autonomy to establish measures as long as they
consider necessary to fight against COVID-19 dissemination. As a result many
States adopted different measures, but the question proposed in italy remains:
Could we expect that the number of daily new cases from one State helps to
explain and predict this number in other State or only the past data of the
same region is enough?
US has 51 States and work with this number of time-series would be useless
and time consuming considering the scope of this research. Thus, we choose the
state with highest number of positive cases (California) and its surrounding
States (Oregon, Nevada, Arizona) also presented in Table 2.
Closer to US policy, in Brazil each State were in charge of defining necessary
measures to avoid COVID-19 dissemination. Here we bring the RJ city health
regions time-series with the same question, but we want to evaluate if the
lower distance betwwen these health regions (comparing to US States and IT
regions distances) could bring us a different result comparing to IT and US
time-series. RJ health regions are also presented in Table 2.
In Figures 3, 4 and 5 we can see that COVID-19 pandemic started at
different dates (presented below) and the data set range of each country or
region can also vary according to data set source. To establish comparisons
between forecasting techniques we work with the same time-series range to all
regions. Thus, in this research we work with time-series range of 369 days.
• Rio de Janeiro city time-series range available: from 13/01/20 to 22/12/21,
but we decide to start on 12/03/20 (when cases start to appear in every
day). 651 days;
• Italy regions time-series range available: from 24/02/20 to 27/07/21. 520
days;
• US time-series range available: from 04/03/20 to 03/07/21. 369 days.
We consider that evaluate all this time series using univariate and multi-
variate approaches because of the high correlation that each time-series has
with the others in the same place at the same lag (correlation) and at differ-
ent lags (auto-correlation) as we can see in Figure 6. All correlation plots are
available in section C.
All data sets are divided into training and test data and their lengths are
341 and 28 days respectively. Our short-term forecasting is 28 days ahead. The
reasons for choosing forecasting range of 28 days are presented below.
 Springer Nature 2021 LATEX template

12 Comparing short-term univariate and multivariate time series forecasting ...

2020−03−12 / 2021−03−15

R1
R2
600 R3 600
R4
R5

400 400

200 200

Mar 12 Mai 01 Jul 01 Set 01 Nov 01 Jan 01 Mar 01

2020 2020 2020 2020 2020 2021 2021

Fig. 3 Rio de Janeiro city COVID-19 daily cases per health region. Source: authors.

2020−02−24 / 2021−02−26

Center
15000 Islands 15000
North−East
North−West
South
10000 10000

5000 5000

Fev 24 Abr 01 Jun 01 Ago 01 Out 01 Dez 01 Fev 01

2020 2020 2020 2020 2020 2020 2021

Fig. 4 Italy regions COVID-19 data per region. Source: authors.

• We work with 341 past observations which is more than 10 times the pre-
diction length. It do not means that our past data is larger enough to well
train some models and give reliable predictions;
• Forecasting daily new cases four weeks ahead allow decision makers in health
units to better plain resource availability or governments to choose ade-
quate measures. At least, better than in most of the previous research daily
predictions worked with shorter forecasting range (seven or fourteen);
• Taking into account that resource availability depends on health units
resource allocation (doctors, beds, among others), the smaller time window
 Springer Nature 2021 LATEX template

Comparing short-term univariate and multivariate time series forecasting ... 13

2020−03−04 / 2021−03−07

50000 Arizona 50000

California
Nevada
40000 40000
Oregon

30000 30000

20000 20000

10000 10000

Mar 04 Mai 01 Jul 01 Set 01 Nov 01 Jan 01 Mar 01

2020 2020 2020 2020 2020 2021 2021

Fig. 5 US states COVID-19 data per region. Source: authors.

unit (until this moment is daily data) we are able to work with and pro-
vide reliable predictions will result in the more useful information to help
decision makers to meet resources requirements that ensure a adequately
treatment to patient demand.

Table 2 Regions or States division (for more details see Appendix B). Source: authors.

RJ (Figure B1) IT (Figure B2) US (Figure B3)

Region (R) 1 Center (CEN) Arizona (AZ)

R2 Islands (ISL) California (CA)
R3 North-West (NOW) Nevada (NV)
R4 North-East (NOE) Oregon (OR)
R5 South (SOT) -

4 Forecasting Models Applied

In this research we expand the framework proposed by Marques et al. (2021) by
using more univariate approaches and adding multivariate approach. Models
applied in next sections are presented in Table 3.
To build each model is necessary estimate some times more than 10 param-
eters and present all of them in 1 or more table is not the aim of this research.
Thus, we briefly explain below the main features of each model.
• ES: ETS is a class of models that essentially works with 2 components
equations trend and season that can be added or multiplied to the remainder.
In each model these components can not be significant, also known as None
(N) or can be significant and better describe original time series features
 Springer Nature 2021 LATEX template

14 Comparing short-term univariate and multivariate time series forecasting ...

Fig. 6 Auto-correlation plot between NV and OR. Source: authors.

Table 3 Forecasting models applied in this research. Source: authors.

Models Univariate Multivariate

CSM ES, ARIMA VAR

CM - -
SSM TBATS, SSM Univariate (SSM-U) SSM Multivariate (SSM-M)
MLM MLP, NNETAR -

as follows: Additive (A) or Additive Damped (Ad) or Multiplicative (M).

This class of models can be combined in 18 different ways (A,N,A; M,Ad,M;
for instance). Equations of each model are presented in Figure 7. For more
details see Hyndman and Athanasopoulos [16];
• ARIMA: ARIMA or Seasonal ARIMA (SARIMA) is a class of models that
combine autoregressive (AR) and moving average (MA) with differenced
values. The AR part of ARIMA (p) shows that the time series is regressed
on its own past data. The MA part of ARIMA (q) indicates that the forecast
error is a linear combination of past respective errors. The I part of ARIMA
(d) shows that the data values have been replaced with differenced values of
d order to obtain stationary data, which is the requirement of the ARIMA
model approach [98]. When we work with SARIMA the same components
appears lagged by the length of seasonal time window (frequency) as P ,
D and Q. For instance, ARIMA (p = 5, d = 0, q = 3)(P = 0, D = 1,
 Springer Nature 2021 LATEX template

Comparing short-term univariate and multivariate time series forecasting ... 15

Fig. 7 ETS equations. Source: Hyndman and Athanasopoulos [16]

Q = 1)[f requency = 7]. For more details see Hyndman and Athanasopoulos
[16] and Kotu and Deshpande [98];
• Space State Model Univariate (SSM-U): The state of a deterministic
dynamic system is the smallest vector that summarises the past of the system
in full [77]. The linearity of state dynamics and observation process, as well as
the normal distribution of noise in state dynamics and measurements are the
assumptions of SSM. A linear autoregressive equation x(t) = A ∗ x(t) + W (t)
where W (t) ≈ N (0, Q) with a measurement that is y(t) = C ∗ y(t) + V (t)
where V (t) ≈ N (0, R), define the linearized process in which y(t) ∈ R. The
random variables W (t) and V (t) represent the process and measurement
noise respectively and are assumed to be independent of each other and with
normal distributions. In our case we will work with a vector length (n = 2
for linear model and n = 3 for order 2 polynomial model) which means a
n ∗ n dimensions A matrix. We select the best approach for each time-series
based on the lowest Akaike Information Criterion (AIC) criteria.
• MLP: MLP is a supplement of feed forward neural network. It consists of
three types of layers—the input layer, output layer and hidden layer. The
input layer receives the input signal to be processed. An arbitrary number of
hidden layers that are placed in between the input and output layer are the
 Springer Nature 2021 LATEX template

16 Comparing short-term univariate and multivariate time series forecasting ...

true computational engine of the MLP. Similar to a feed forward network in

a MLP the data flows in the forward direction from input to output layer.
The neurons in the MLP are trained with the back propagation learning
algorithm. MLPs are designed to approximate any continuous function and
can solve problems which are not linearly separable. In time-series problem
the input layer is past observations and we set then to choose between 1 and
28 (prediction length) according to Mean Square Error the optimal number
of lags used and which lags will be used. The same criteria were used to
define number of hidden nodes in each hidden layer;
• NNETAR: NNETAR is a feed-forward neural networks with a single hid-
den layer and lagged inputs. This model works with 2 (for non seasonal
time-series) or 3 (for seasonal time-series) parameters: the number of past
observations used as input layers (p), the number of past observations lagged
by the length of seasonal time window used as input layers (P ) and the
number of neurons (k) in the single layer. For instance, (p = 21, P = 1,
k = 11)[7]. For more details see Hyndman and Athanasopoulos [16];
• TBATS: BATS model is Exponential Smoothing Method + Box-Cox Trans-
formation + ARMA model for residuals. Aiming to reduced the parameters
of model when the frequencies of seasonalities are high and giving more flexi-
bility to deal with complex seasonality, De Livera et al. [99] propsed TBATS
model which is BATS model + Trigonometric Seasonal. Equations TBATS
model is presented in equations below where ω and ϕ are Box-Cox and the
damping parameters respectively, ARMA(p, q) process model the error and
m1 to mJ list the seasonal periods used (in our case there is only m1 always
equal to 7) while k1 to kJ are the corresponding number of Fourier terms
used (in our case there is only k1 ). For instance, TBATS (ω = 0.21, [p = 0,
q = 0], ϕ = 0.96, [< m1 = 7, k1 = 3 >]).

(ω)
(ω) yt −1
yt = , ω ̸= 0,
ω
(ω)
yt = log yt , ω = 0,

t
(ω)
X
yt = lt−1 + ϕ ∗ bt−1 + sit−mi + dt ,
i=1

lt = lt−1 + ϕ ∗ bt−1 + α ∗ dt ,

bt = (1 − ϕ) ∗ bt + ϕ ∗ bt−1 + β ∗ dt ,

sit = sit−mi + γi ∗ dt ,
p
X q
X
dt = ϕi ∗ dt−i + θi ∗ ϵt−i + ϵt ,
i=1 i=1
 Springer Nature 2021 LATEX template

Comparing short-term univariate and multivariate time series forecasting ... 17

kj
X
sit = sij,t ,
j=1

sit = sij,t−1 ∗ cos λij + s∗i i i

j,t−1 ∗ sin λj + γ1 ∗ dt ,

s∗i i i ∗i i i
t = sj,t−1 ∗ sin λj + sj,t−1 ∗ cos λj + γ2 ∗ dt

• VAR: A VAR(p) model is a generalisation of the univariate autoregressive

(AR) where (p) shows that the time series is regressed on past data of all
time-series for forecasting a vector of time series [16]. Each variable has
one equation that includes a constant and lags of all of the variables in the
system.
• Space State Model Multivariate (SSM-M): A SSM-M model is a generali-
sation of SSM-U and works similarly, but with y(t) ∈ Rm where m is the
number of time-series considered.
We choose the Root Mean Square Error (RMSE) accuracy criteria to com-
pare all models presented in Table 3. The results of each model is presented in
Tables 4, 5 and 6. The error evaluation is divided in 3 parts and was applied
to using each model:
• In-sample (RMSE IN): comparing training data with fitted values obtained;
• Out-sample all (RMSE OUT-ALL): comparing all test data with predicted
values obtained;
• Out-sample mean (RMSE OUT-MEAN): comparing a piece of test data (7
days ahead) with predicted values obtained 4 times and calculate the average
error. We run the same model without parameters re-estimation, but we add
a new week data (7 days).

4.1 Univariate Approaches

In this section we apply the methods presented in the second column of Table
4 to each health region (Rio de Janeiro city), region (Italy) and states (US).
Tables 4, 5 and 6 summarizes the models obtained by each method and in
Tables 7, 8 and 9 the RMSE IN, RMSE OUT-ALL and RMSE OUT-MEAN
are presented for each model. These methods only consider the previous values
of the same variable to make predictions. In all models the seasonality time
window is 7 (weekly).
All plots of time-series approach applied are available in supplementary
material. In Figure 8 an example is provided showing the results of ETS model
application in Center (Italy Region). The best type model (ETS(M,Ad,M)) in
each class of model (ETS) for Figure 8, for instance is chosen by the lowest
AIC criteria.
 Springer Nature 2021 LATEX template

18 Comparing short-term univariate and multivariate time series forecasting ...

Center Time-serie evaluation: ETS(M,Ad,M)

7000

training data
fitted values
5000

test data
prediction
Daily cases

3000
1000
0

0 100 200 300

Days

Fig. 8 ETS model applied in Center. Source: The authors.

Table 4 Forecasting parameters per model per region for Rio de Janeiro health region. Source: The authors.

Models R1 R2 R3 R4 R5

ES A,N,A A,N,A A,N,A A,N,A A,N,A

ARIMA (5,0,3)(0,1,1)[7] (2,0,1)(0,1,2)[7] (2,0,1)(0,1,2)[7] (2,0,1)(0,1,2)[7] (2,0,1)(0,1,2)[7]
SSM-U 3 3 3 3 3
12 inputs, 8 11 inputs, 8 12 inputs, 9 10 inputs, 8 10 inputs, 8
MLP
hidden nodes hidden nodes hidden nodes hidden nodes hidden nodes
NNETAR (21,1,11)[7] (14,1,8)[7] (15,1,8)[7] (6,1,4)[7] (15,1,8)[7]
TBATS 1, {4,2}, -, {<7,3>} 1, {4,2}, -, {<7,3>} 1, {2,2}, -, {<7,3>} 1, {4,4}, -, {<7,2>} 1, {4,2}, 0.865, {<7,3>}

Table 5 Forecasting parameters per model per region for Italy regions. Source: The authors.

Models CEN ISL NOW NOE SOT

ES M,Ad,M A,N,A M,A,M M,N,M A,Ad,A

ARIMA (1,0,1)(0,1,2)[7] (2,1,2)(0,0,0)[7] (1,0,1)(0,1,2)[7] (4,0,3)(0,1,2)[7] (5,1,2)(1,0,0)[7]
SSM-U 3 3 3 3 3
16 inputs, 12 12 inputs, 10 17 inputs, 12 16 inputs, 12 13 inputs, 10
MLP
hidden nodes hidden nodes hidden nodes hidden nodes hidden nodes
NNETAR (23,1,12)[7] (25,1,13)[7] (21,1,11)[7] (25,1,13)[7] (23,1,12)[7]
TBATS 0.21, {0,0}, 0.96, {<7,2>} 1, {0,0}, -, {<7,2>} 0.25, {0,0}, 0.96, {<7,3>} 0.07, {0,0}, 1, {<7,2>} 1, {4,3}, 0.89, {<7,2>}
Springer Nature 2021 LATEX template
Table 6 Forecasting parameters per model per state for US. Source: The authors.

Models AZ CA NV OR

ES A,Ad,A A,N,A A,N,A A,N,A

ARIMA (0,1,2)(0,0,2)[7] (0,1,2)(0,0,1)[7] (0,1,1)(0,0,2)[7] (2,1,3)(0,0,2)[7]
SSM-U 2 3 3 3
21 inputs, 14
12 inputs, 10 13 inputs, 9 17 inputs, 10
MLP hidden nodes
hidden nodes hidden nodes hidden nodes
Difference order:1
NNETAR (22,1,12)[7] (22,1,12)[7] (25,1,13)[7] (25,1,13)[7]
TBATS 1, {0,0}, 0.8, {<7,3>} 1, {3,3}, -, - 1, {0,0}, -, {<7,2>} 1, {0,0}, -, {<7,1>}

Table 7 Forecasting results per model per region for Rio de Janeiro health region. Source: The authors.
20

RMSE IN RMSE OUT-ALL RMSE OUT-MEAN

Models
R1 R2 R3 R4 R5 R1 R2 R3 R4 R5 R1 R2 R3 R4 R5

ES 18.6 62.5 77.2 52.7 73.2 19.9 57.2 119.4 47 104.9 14.4 43.5 65.8 33.9 64
ARIMA 17 59.5 73.1 49.7 67.4 16.6 51.5 99.5 37.9 112 13.5 46.4 69.9 38.1 63.2
SSM-U 20.4 72.5 89.5 57.4 83.5 24.8 75.3 137.7 58.9 118.5 19.5 66.7 107.3 48.1 82.1
MLP 8.8 31.3 31.7 26.2 29.8 17.4 53 92.5 34.3 91.7 15.2 45.9 79.8 38.5 62.6
NNETAR 2.1 21.8 23.4 38.8 21 16.5 65.4 99.4 46.4 101.4 13.2 66.1 96 48 66.5
TBATS 17.2 59.5 72 51.7 72.7 20.4 56.3 118.1 49 104.1 14.1 46.6 78.4 39.5 54.9
Springer Nature 2021 LATEX template

Comparing short-term univariate

Table 8 Forecasting results per model per region for Italy regions. Source: The authors.

RMSE IN RMSE OUT-ALL RMSE OUT-MEAN

Models
CEN ISL NOW NOE SOT CEN ISL NOW NOE SOT CEN ISL NOW NOE SOT

ES 256.3 128.4 1085.9 757.8 364.7 436.7 454.3 827.7 991.9 614 414.2 143.4 764.6 673.7 368.8
ARIMA 242.8 127.1 429.9 761 348.8 418.2 489.8 611.4 974.9 632.7 382.1 155.5 676.1 914.1 541.1
SSM-U 165.2 54 397 587.4 238.8 551.4 420.4 493.1 1237.5 841.7 390.3 132.1 500.1 713.3 550.4
MLP 73.9 38.6 103.5 242.7 81.1 471.2 879.7 1064.5 1424.7 1002.3 438.9 406.8 951.2 938.4 681.7
NNETAR 36.9 17.9 71 77.9 45.7 544.3 426.4 2717.5 1720.3 1386.8 404.9 153.9 1474.5 1086.9 609.3
TBATS 238.4 127 427.9 782.7 329.5 364.9 463.2 542.1 674.7 488.4 355.3 145.1 498.3 739.3 494.2

Table 9 Forecasting results per model per state for US. Source: The authors.

RMSE IN RMSE OUT-ALL RMSE OUT-MEAN

Models
AZ CA NV OR AZ CA NV OR AZ CA NV OR

ES 1265.8 3167.4 274.3 152.3 662.5 7774.3 389.8 304.5 557.2 1996.3 194.5 152.5
ARIMA 1286.4 2980.6 276.1 153.5 1195.4 6838 334.4 324.1 669.8 1959.1 190.1 148.9
SSM-U 1165.6 1616.6 193.2 113.3 1568.3 7609.9 381.1 274.3 779.8 2301.4 219.8 145.5
MLP 229 1210.5 102.7 48.7 1749.9 9550.3 150.7 490.7 843.9 2507.3 149 255.7
NNETAR 152.1 500.8 37.1 18.5 948.4 29.3x103 234 694 699 7023.6 182.1 318.1
TBATS 1251.7 2913.2 275.3 152.1 617.7 9332.4 395.3 301 590 2617 197.2 153.6
Springer Nature 2021 LATEX template
 Springer Nature 2021 LATEX template

22 Comparing short-term univariate and multivariate time series forecasting ...

From Tables 7, 8 and 9 we can conclude that the best error in-sample
considering RMSE criteria are NNETAR (with thirteen) and MLP (with one)
for all time-series which is not surprising since neural networks work better
the more data we give them.
Although outperforming on in-sample comparison, ML models do not
obtained the same result by evaluating RMSE OUT-ALL and RMSE OUT-
MEAN in which they got lowest RMSE in only five and two time series
respectively.
Trying to predict daily cases 28 days ahead without add new data or param-
eter re-estimation (OUT-ALL), MLP showed better results for four RJ health
regions and one US State. In second place, TBATS showed better results for
Three IT regions and one US State. SSM-U appeared in third position being
chosen in two IT regions and one US State.
However, when we predict daily cases 28 days ahead adding new data
weekly without parameter re-estimation (OUT-MEAN) we conclude that ES
models give us better predictions for six time-series while TBATS models and
SMM-U were chosen for three and two time-series respectively. All these results
are summarized in Table 10.
SSM-U best approach considering the lowest AIC criteria were order two
polynomial model (n = 3) in thirteen time-series. Only in AZ time-series linear
model (n = 2) was chosen.

Table 10 Best performance models frequency. Source: The authors.

RMSE IN RMSE OUT-ALL RMSE OUT-MEAN

Models
RJ IT US Total RJ IT US Total RJ IT US Total

ES 0 0 0 0 0 0 0 0 3 2 1 6
ARIMA 0 0 0 0 0 0 1 1 0 0 1 1
SMM-U 0 0 0 0 0 2 1 3 0 1 1 2
MLP 1 0 0 1 4 0 1 5 0 0 1 1
NNETAR 4 5 4 13 1 0 0 1 1 0 0 1
TBATS 0 0 0 0 0 3 1 4 1 2 0 3

After comparing 6 different class of univariate forecasting models and point

out which class of model according to lowest RMSE criteria, in next section
we present two multivariate forecasting models.

4.2 Multivariate Approaches

In this section we apply the methods presented in the third column of Table
3 (SSM-M and VAR) to each health region (RJ), region (IT) and states (US).
These methods consider previous values of all variables available to make
predictions. In Table 11 we summarize the forecasting error results.
VAR models are divided in four types of deterministic regressors: none,
constant, trend or both (constant and trend). We select the deterministic
 Springer Nature 2021 LATEX template

Comparing short-term univariate and multivariate time series forecasting ... 23

regressors type to each multivariate time-series using the lowest AIC criteria.
In addition, to select the VAR model order (p) we adopt the Schwarz Crite-
rion (SC(n)) and obtained p = 1 to RJ with constant and trend deterministic
regressors (18 parameters), p = 23 to IT with trend deterministic regressors
(113 parameters) and p = 2 to US with constant and trend deterministic
regressors (20 parameters).
In SSM-M we select a vector length (nx) that gave us the lowest error con-
sidering Akaike Information Criterion (AIC). The nx can be 8 (linear model)
or 12 (polynomial order 2 model) to US and 10 (linear model) or 15 (polyno-
mial order 2 model) to RJ and IT (two or three times the number of univariate
time series).
Table 11 Forecasting results per multivariate model per region for Rio de Janeiro, Italy, and US. Source: The authors.

RMSE IN RMSE OUT-ALL RMSE OUT-MEAN

Models
RJ R1 R2 R3 R4 R5 R1 R2 R3 R4 R5 R1 R2 R3 R4 R5

VAR(1) 102.8 67.7 110.3 55.7 83.0 63.3 100.3 166.8 66.7 118.5 29.4 58.7 94.9 46.8 71.1
SSM-M(10) 21.2 73 91.3 59.5 86.4 83.1 107.7 184.2 78.4 123.8 20.0 68.9 112.3 45.5 79.5

IT CEN ISL NOW NOE SOT CEN ISL NOW NOE SOT CEN ISL NOW NOE SOT

VAR(23) 2331.1 3686.2 2292.1 308.3 2322.3 4749.5 1533.5 4522 11x103 4502.2 1026.5 1024.2 1326.4 2163.3 1062.1
SSM-M(10) 216.6 94.1 477.7 742.8 335.2 2080.8 3389.4 1808 2022.5 1824.3 589.7 262.5 833.9 1231.9 565.0

US AZ CA NV OR - AZ CA NV OR - AZ CA NV OR -

VAR(2) 2801.1 14.9x103 559.5 142.9 - 2908.8 13.1x103 992.9 461.4 - 959.9 3519.2 394.1 185.0 -
SSM-M(8) 1341.2 2047.9 229.8 131.6 - 3860.8 4676.9 4645.5 4674.9 - 991.7 785.1 344.8 115.7 -
24
Springer Nature 2021 LATEX template

Comparing short-term univariate

 Springer Nature 2021 LATEX template

Comparing short-term univariate and multivariate time series forecasting ... 25

From Table 11 we can conclude that:

• Linear models were chosen for all RJ (n=10), IT (n=10) and US (n=10) data
considering the lowest AIC criteria. In univariate time-series we obtained
the opposite (almost all models obtained lowest AIC with polynomial order
2 models).
• VAR(1) model obtained best in-sample (IN) error in four RJ health regions
while in three US States and four IT regions SSM-M(8) and SSM-M(10)
outperform VAR approach considering in-sample error. In other words, SSM-
M models better fitted training data in ten time-series training while VAR
models better fitted the other four time-series;
• Predicting 28 days ahead without add new data or parameter re-estimation
(OUT-ALL) VAR models tied with SSM-M models. VAR(2) achieve better
results in three US States while SSM-M(10) fitted better four IT regions.
• predicting 28 days ahead adding new data weekly without parameter re-
estimation (OUT-MEAN) SSM-M models better fitted all IT regions and
two US States while VAR(1) better fitted four RJ health regions. In other
words, SSM-M models showed better results for eight time-series while VAR
models better fitted the other six time-series.
In the next section we compare results obtained with all approaches men-
tioned in Table 3 and detailed presented in Tables 4, 5, 6, 7, 8, 9 and
11.

4.3 Comparing results of Univariate and Multivariate

methods
In Table 12 we compare best model (univariate and multivariate) for all
time-series considering IN, OUT-ALL and OUT-MEAN RMSE results. This
comparison combines Tables 7, 8, 9 and 11 presented in previous sections.
Table 12 Univariate and Multivariate selected models and error. Source: The authors

IN OUT-ALL OUT-MEAN

Univariate Multivariate Univariate Multivariate Univariate Multivariate

Data Regions Model RMSE Model RMSE Model RMSE Model RMSE Model RMSE Model RMSE

R 1 NNETAR 2.1 SSM-M 21.2 NNETAR 16.5 SSM-M 93.3 NNETAR 13.2 SSM-M 20
R 2 NNETAR 21.8 VAR 70.1 ARIMA 51.5 VAR 90.9 ES 43.5 VAR 56.9
R 3 NNETAR 23.4 SSM-M 91.3 MLP 92.5 VAR 92.2 ES 65.8 VAR 86
RJ
R 4 MLP 26.2 VAR 59.4 MLP 34.3 SSM-M 84.1 ES 33.9 VAR 42.2
R 5 NNETAR 21 VAR 85.6 MLP 91.7 VAR 92.1 TBATS 54.9 VAR 69.6

CEN NNETAR 37.6 SSM-M 216.6 TBATS 364.9 SSM-M 2104.3 TBATS 355.3 SSM-M 589.7
ISL NNETAR 17.2 SSM-M 94.1 ES 454.3 VAR 1496.7 ES 143.4 SSM-M 262.5
Italy
NOW NNETAR 73.4 SSM-M 477.7 TBATS 542.1 SSM-M 1843.5 TBATS 498.3 SSM-M 833.9
NOE NNETAR 78.3 VAR 308.3 TBATS 674.7 SSM-M 2049.1 ES 673.7 SSM-M 1231.9
SOT NNETAR 45.1 SSM-M 335.2 TBATS 488.4 SSM-M 1860.5 ES 368.7 SSM-M 565.0

AZ NNETAR 152.1 SSM-M 1341.2 TBATS 617.7 VAR 1986 ES 557.2 VAR 900
CA NNETAR 513.3 SSM-M 2047.9 MLP 164.9 SSM-M 5643.6 ARIMA 1959.1 SSM-M 785.1
US
NV NNETAR 37.6 SSM-M 229.8 NNETAR 246.7 VAR 406.5 MLP 157.2 VAR 298.1
OR NNETAR 17.7 SSM-M 131.6 SSM-M 268.6 VAR 117.4 ARIMA 148.9 SSM-M 115.7
26
Springer Nature 2021 LATEX template

Comparing short-term univariate

 Springer Nature 2021 LATEX template

Comparing short-term univariate and multivariate time series forecasting ... 27

Table 12 reinforces the flexibility of neural networks to fit training data

when working with a large number of observations. Results obtained by
NNETAR models outperform all univariate and multivariate models applied
in this research. Although NNETAR not shows the same performance taking
into account out-sample results, we see another neural network method (MLP)
providing the lowest RMSE OUT-ALL to 4 of 5 RJ health regions.
Despite of high correlation between variables of RJ, IT and US (see Figure
6 and supplementary data) time-series data we see multivariate approach out-
perform only in R3 and R5 from RJ and in OR from US (OUT-ALL) using
VAR models and OR from US (OUT-MEAN) using SSM-M.
It is important to emphasize that, although univariate models obtained
the lowest RMSE in 38 of 42 time-series, the difference of results between
univariate and multivariate best approaches is lower in RJ than in IT and
US. It may occur because health regions in RJ city are close comparing to US
states or IT regions.
The univariate methods could also outperform multivariate because we
chose pure simpler models (SSM-M and VAR) and we did not combine them
or propose to include more complex models on this analysis like VARMA or
some neural network multivariate method.

4.4 Forecasting 28 days ahead

In Table 12 we compared the error results between univariate and multivariate
approach which provide us many useful insights.
In this section we summarize the results presented in Table 12 to select
best model for each time-series evaluated and then apply it considering all data
available (training and test data) to predict daily new cases 28 days ahead.
To provide the daily new cases prediction proposed we re-estimate all
parameters of models selected in third column of Table 13. Finally, in Figures
9, 10 and 11 we present the forecasting values with confidence interval of 0.95.
 Springer Nature 2021 LATEX template

28 Comparing short-term univariate and multivariate time series forecasting ...

1 ARIMA(1,0,1)(1,1,2)[7] 2 ARIMA(2,0,1)(0,1,2)[7] 3 MLP − HD 8 Diff order:

600

600
150

400
Daily cases

Daily cases

Daily cases

400
100

200

200
50
0

0
0 100 200 300 400 0 100 200 300 400 0 100 200 300 400

Days Days Days

4 MLP − HD 9 Diff order: 5 MLP − HD 9 Diff order:

800
600
400
Daily cases

Daily cases

400
200

200
0

0 100 200 300 400 0 100 200 300 400

Days Days

Fig. 9 RJ Health Regions forecasting. Source: The authors.

Center TBATS(0.215, {0,0}, 0.961, {<7,2>}) Islands SSM−M nx: 10 North−East SSM−M nx: 10
2000 4000 6000

6000
1500
Daily cases

Daily cases

Daily cases

2000
500
0

0 100 200 300 400 0 100 200 300 400 0 100 200 300 400

Days Days Days

North−West TBATS(0.156, {0,0}, 1, {<7,2>}) South TBATS(1, {3,3}, 1, {<7,2>})

15000

6000
Daily cases

Daily cases
5000

2000
0
0

0 100 200 300 400 0 100 200 300 400

Days Days

Fig. 10 IT Regions forecasting. Source: The authors.

AZ TBATS(1, {0,0}, 0.801, {<7,3>}) CA SSM-M nx: 8

Daily cases

Daily cases
10000

30000
0

0 100 200 300 400 0 100 200 300 400

Days Days

NV MLP - HD 10 Diff order: OR SSM-U nx: 3

3500
Daily cases

Daily cases

1000
1500
0

0 100 200 300 400 0 100 200 300 400

Days Days

Fig. 11 US States forecasting. Source: The authors.

 Springer Nature 2021 LATEX template

Comparing short-term univariate and multivariate time series forecasting ... 29

Table 13 Best model selection to each time-series. Source: The authors.

Region/State RMSE IN RMSE OUT-ALL RMSE OUT-MEAN

R1 NNETAR ARIMA ARIMA

R2 NNETAR ARIMA ETS
R3 NNETAR MLP ETS
R4 MLP MLP ETS
R5 NNETAR MLP TBATS

CEN NNETAR TBATS TBATS

ISL NNETAR SSM-U SSM-U
NOW NNETAR SSM-U TBATS
NOE NNETAR TBATS ETS
SOT NNETAR TBATS ETS

AZ NNETAR TBATS ETS

CA NNETAR SSM-M SSM-M
NV NNETAR MLP MLP
OR NNETAR SSM-U SSM-M

5 Conclusions
In this research we applied 6 univariate and 2 multivariate models in order
to find which approach best fit and predict 14 time-series from Brazilian city
(RJ), All Italian regions and 4 US States.
We chose to apply multivariate methods which is unusual in current liter-
ature of human infectious disease outbreak prediction or forecasting (less then
10% of research we found) considering the high correlation and auto-correlation
between different time-series from the same place in many lags as we saw in
Figure 6. In Appendix C all auto-correlation plots are presented where we see
a significant correlation between regions data until lag 15 to RJ and in all lags
to Italy regions and US States.
Besides the strong potential of multivariate methods, we did not observe
them outperforming univariate methods.
In-sample results obtained best results using ML methods for all time-series
which is expected considering that these type of models usually provide better
results the more data is available for training. However, the same result was
not observed in out-sample results evaluation.
Our prediction presented in Figures 9, 10 and 11, suggests that in the next
28 days:
• 4 RJ health regions will remain on the same level of daily new cases, but
in R5 is expected to face a considerable increasing of daily COVID-19 new
cases. However, it will be at least lower than levels observed in previous data;
• IT regions will face a exponential increasing of daily COVID-19 new cases,
excluding CEN Region;
• In US States we can expect different behaviours of daily COVID-19 new
cases. To AZ it is expected a tiny decreasing while in CA and NV will
 Springer Nature 2021 LATEX template

30 Comparing short-term univariate and multivariate time series forecasting ...

increase. In OR it is expected that daily cases remains in the same level of

600 new daily cases.
As further research, we suggest the application of multivariate MLM tech-
niques like multivariate MLP or LSTM (largely and successfully applied in
literature for univariate time-series approach). Another possible way is to
combine the mentioned multivariate methods with VARMA.
Causal models are largely applied in current literature and should be also
explored. However, this type of approach also depends from collecting data
from other sources that sometimes unavailable.
In airborne infectious diseases transmission like COVID-19, ILI, among
others we observe interesting applications combining daily/weekly or monthly
cases with search engine of Google or Baidu (in China) or mobility data to find
better predictions. On the other hand, diseases transmitted by vectors such as
mosquitoes like Dengue, Zika-Virus among others are typically combined with
temperature and rainfalls, for instance.

Supplementary information. The article has accompanying supplemen-

tary.

Declarations
• Funding: The author(s) received no specific funding for this work.
• Conflict of interest/Competing interests: The authors have declared that no
competing interests exist.
• Ethics approval: Not applicable
• Consent to participate: Not applicable
• Consent for publication: Not applicable
• Availability of data and materials: All relevant data are within the
manuscript and its Supporting Information files.
• Code availability: https://github.com/DanielAssad/Short-term-forecasting.
git

Appendix A Research synthesis

In this section we present all similar research that we found in literature. As
mentioned in Section 2.5, columns 3 to 7 address the approach used in each
research between CSM, CM, SSM and MLM.
In columns 8 to 9 is presented the range of time windows used as well as
the prediction range. Time windows found on previous research were day (d),
week (w), month (m) or year (y). The prediction range are expressed in time
windows mentioned, but some models proposed to forecast the whole pandemic
period (wpp).
 Springer Nature 2021 LATEX template

Comparing short-term univariate and multivariate time series forecasting ... 31

The variable measured/evaluated and forecasted in each research (column

10) on those time windows are number of patient cases (ca), deaths (de) and
recovered (re), admitted and discharged from hospital or intensive care unit
(adhosp and dishosp) and transmission rate (rt).
Excluding rt, all measures mentioned can be counted in two different ways:
by time window or cumulative. For example, daily cases (dca), monthly deaths
(mde), yearly patients admitted in hospital (yadhosp), cumulative cases (Cca),
cumulative patients discharged from hospital (Cdishosp).
Columns 11 to 13 presents in which countries each research applied the
methods specified in columns 3 to 7, the type of forecast approach divided
in Univariate (Uni), Causal or Multivariate (Mul) and the disease outbreak
studied.
 Table A1: Literature Review Synthesis. Source: The authors.

Data
pre- Disease
Authors Year CSM CM SSM MLM Others Data Range Measure Country Approach
dic- outbreak
tion

Brazil, Uni
ES, KF, MLP, 24/02/20 to
Current 2022 - - 28 d dca Italy, and COVID-19
ARIMA TBATS NNETAR 11/07/21
US Mul
Spatial
auto- 21/01/20 to
[12] 2021 - SEIR - - wpp dca China Causal COVID-19
corre- 12/03/20
lation
01/01/10 to
[26] 2021 ARIMA - - - - 6m mca Thailand Causal Dengue
12/12/16
Cca, 16
22/01/20 to
[13] 2021 ARIMA - - LSTM - 60 d Cde, coun- Uni COVID-19
12/03/ 20
32

Cre tries
NNETAR,
ES, 20/02/20 to Cca,
[27] 2021 - TBATS ELM, Prophet 30 d Iran Uni COVID-19
ARIMA 15/08/20 Cde
MLP
Newbolt-
ES, Granger, 21/01/20 to
[14] 2021 SIR - RNN 4w rt Greece Uni COVID-19
ARIMA Bates- 14/08/ 20
Granger
Springer Nature 2021 LATEX template

Comparing short-term univariate

 Table A1 continued from previous page
10
GML, 22/01/20 to dca,
[60] 2021 - - - - wpp coun- Uni COVID-19
GMR 14/09/ 20 rt
tries
15/03/20 to dca, Brazil
[78] 2021 - - KF AutoML - 15 d Mul COVID-19
17/05/20 dde (Ceará)
Brazil
01/01/07 to
[84] 2020 - - - EEMD - 3m mca (5 Mul Meningitis
31/12/18
states)
Spain,
dca,
Sierra
dde, COVID-19,
GML, Gaussian Leone
[15] 2020 - - - 65 to 104 d 10 d dad- Uni Ebola, and
GMR Model and
hosp, zika virus
Colom-
dishosp
bia
dca,
08/03/20 to
[28] 2020 VAR - - - - 50 d dde, Iran Mul COVID-19
27/06/20
dre
01/03/10 to 2 to Japan
[85] 2020 - - - CLAGNN- wca Mul ILI
30/03/19 15 w and US
22/01/20 to 5 coun-
[61] 2020 - GML - RNN Prophet 6m Cca Uni COVID-19
16/07/ 20 tries
ARIMA, Brazil
[29] 2020 ARI- - - RNN - 2007 to 2015 18 w wca (Fort- Causal Dengue
MAX aleza)
Springer Nature 2021 LATEX template
 Table A1 continued from previous page
China
SVM,
[30] 2020 ARIMA - - - 2004 to 2016 24 m mca (31 Causal Tuberculosis
BMA
cities)
2006 and Brazil
[31] 2020 ARIMA - - - - 36 m mca Uni Leprosy
2016 (Maranhão)
[32] 2019 ARIMA - - SVM - 2004 to 2017 6m mca China Causal HFRS
LSTM, 01/06/08 to
[33] 2019 ARIMA - - - 18 m mca China Uni HFMD
NARNN 30/06/18
01/01/11 to
[34] 2019 ARIMAX- - - - 15 m mca China Causal HIV/AIDS
30/01/16
Linear
[35] 2019 ARIMA - - RNN Regres- 2010 to 2018 26 w wca US Causal ILI
sion
UK
wca
[62] 2019 - SEIR - - - 52 weeks 52 w (New Uni Measles
and rt
Castle)
193
1 to
34

[63] 2019 - SIR KF - - 1980 to 2014 yca coun- Uni Measles

50 y
tries
6 w
01/05/90 to
[36] 2019 ARIMA - - NNAR - and 1 mca PhilippinesUni Dengue
31/10/11
y
Brazil
[86] 2019 - - - SVM - 2001 to 2012 4y wca (7 Causal Dengue
cities)
Springer Nature 2021 LATEX template

Comparing short-term univariate

 Table A1 continued from previous page
ARIMA,
[37] 2019 - - - - 2010 to 2017 12 m mca Ukraine Uni Measles
ES
2012 to 2013
[38] 2018 ARIMAX- - - Cluster 4w wca Thailand Causal Malaria
(99 w)
01/01/07 to
[39] 2018 ARIMA - - - Cluster 6m mca China Causal HFRS
31/12/16
1 w 15 to 22 Brazil
[64] 2018 - SEIR - - - 6m wca Uni Zika virus
w 16 (Bahia)
RNN, US and
[40] 2018 ARIMA,VAR
- - - 2009 to 2016 8w wca Mul ILI
CNN Japan
Spain
[65] 2018 - SEIR - - - 2010 to 2014 2w wca (Cat- Causal ILI
alonia)
Space-
time
Autore-
gres-
sive
Inte- China
[41] 2018 ARIMA - - - 2005 to 2014 12 m mca Mul HFRS
grated (Hubei)
Mov-
ing
Aver-
age
(STARIMA)
Springer Nature 2021 LATEX template
 Table A1 continued from previous page
Sequential
Monte
Carlo
approach Canada
[66] 2018 - SEIR - - 1921 to 1956 1 m mca Uni Measles
of par- (Saskatchewan)
ticle
filter-
ing
Colombia
[42] 2017 ARIMA - - - - 2009 to 2012 18 w wca Uni ARI
(Bogota)
US and Dengue
[43] 2017 ARIMA - - KCDE - 1990 to 2010 26 w wca Uni
Japan and ILI
Agent
based
[100] 2017 - - - - 2014 4w wca Liberia Uni Ebola
simu-
lation
Distributed
36

lag
non-
[101] 2017 - - - - 1998 to 2015 6w wca Taiwan Causal Dengue
linear
model
(DLNM)
Springer Nature 2021 LATEX template

Comparing short-term univariate

 Table A1 continued from previous page
Bootstrap
Aggre-
gating
(Bag-
[102] 2017 - - - - ging) 2011 to 2014 12 m mca China Causal ILI
and
Rank
Aggre-
gation
Dengue
2007, 2014 1 to 4 and
[87] 2017 - - - KRRGPN- wca Colombia Uni
to 2016 w Chikun-
gunya
China
(10
[88] 2017 - - - NFIS Cluster 2008 to 2014 12 m mca Causal Dengue
loca-
tions)
ARIMA,
Indonesia
[44] 2016 ARI- - - - - 2010 to 2014 26 w wca Causal Dengue
(Surabaya)
MAX
[45] 2016 ES - - - - 2006 to 2016 6m mca China Uni HFRS
Guinea,
Sierra
[67] 2016 - GM - - - 2014 to 2015 wpp wca Leone Causal Ebola
and
Liberia
Springer Nature 2021 LATEX template
 Table A1 continued from previous page
China
[46] 2016 ARIMA - - - - 1992 to 2013 12 m mca (Jiao- Causal HFRS
nan)
Mexico
1 to 6
[47] 2016 ARIMA - - - - 1985 to 2012 mca (all Causal Dengue
m
states)
Segmented
India
[48] 2016 ES - - - regres- 2014 to 2016 24 m mca Causal Malaria
(Odisha)
sion
Vietnam
(Hanoi
[89] 2016 - - - RF - 2001 to 2012 30 d dca – 29 Mul Cholera
dis-
tricts)
NARNN, China
[49] 2015 ARIMA - - - 2004 to 2012 12 m mca Uni HFRS
GRNN (Jiangsu
Thailand
38

[50] 2015 ARIMA - - - - 2007 to 2013 24 m mca (Chon- Uni DHF

buri)
[51] 2014 ARIMA - - RF - 2006 1w wca Egypt Uni ILI
[52] 2014 ARIMA - - - - 2006 to 2012 26 w wadhosp China Causal HFMD
US
[68] 2014 . SIRS KF - - 2003 to 2012 6w wca (115 Uni ILI
cities)
Springer Nature 2021 LATEX template

Comparing short-term univariate

 Table A1 continued from previous page
US
[69] 2013 - SEIR MCMC - - 2004 to 2006 24 m mca (Ari- Uni SSTIS
zona)
[79] 2013 - - HMM - - 2008 to 2012 26 w wca Portugal Uni ILI
[70] 2012 - SEIR - - - 2009 wpp dadhosp Chile Uni ILI
Thailand
(Chi- Dengue
[71] 2011 - SEIR - - - 1983 to 2006 19 y mca Uni
ang and DHF
Mai)
SIS, Brazil
[59] 2011 - SIR, - GA - 2007 52 w wca (Rio de Uni Dengue
CA Janeiro)
India
(Gujarat
[72] 2010 - SEIR - - - 1987 to 2006 3m mca and Causal Malaria
Rajasthan
states)
China
[53] 2010 ARIMA - - - - 2005 to 2008 1w wca (Hong Causal ILI
Kong)
Brazil
(Sal-
[73] 2009 - CA - - - 1999 5y wca vador Causal Dengue
and
Mossoró)
[90] 2008 - - - WNN - 2002 to 2004 52 w wca US Uni ILI
Springer Nature 2021 LATEX template
 Table A1 continued from previous page
China
[54] 2008 ARIMA - - - - 1957 to 2001 24 m mca (Bei- Uni SF
jing)
ES, Mali ARI and
[55] 2007 - - - - 1996 to 2004 3m mca Uni
ARIMA (Niono) Malaria
[56] 2007 ES - - - - 2002 to 2003 1d dca US Uni ILI
[74] 2005 - SEIR - - - 1984 to 2002 4w wca France Uni ILI
[57] 2001 ARIMA - - - - 1980 to 1995 6m mca US Uni Malaria
[80] 1991 - - MCMC - - 1987 to 1988 36 m mca US Uni HIV/AIDS
[75] 1991 - GM - - - 1985 to 1988 6m mca Brazil Uni HIV/AIDS
40
Springer Nature 2021 LATEX template

Comparing short-term univariate

 Springer Nature 2021 LATEX template

Comparing short-term univariate and multivariate time series forecasting ... 41

Appendix B Geographical Regions

In this section we present the map of regions mentioned in Table 2 for each
time-series evaluated in this research.

Fig. B1 Rio de Janeiro Health Regions. Source: authors.

 Springer Nature 2021 LATEX template

42 Comparing short-term univariate and multivariate time series forecasting ...

Fig. B2 Italy Regions. Source: authors.

 Springer Nature 2021 LATEX template

Comparing short-term univariate and multivariate time series forecasting ... 43

Fig. B3 US States. Source: authors.

 Springer Nature 2021 LATEX template

44 Comparing short-term univariate and multivariate time series forecasting ...

Appendix C Correlation plots

In this section we present the correlation plots between all variables from each
country or region evaluated in this research. In section 3 we present a single
plot in order to show why work with multivariate approach could be worth.

Fig. C4 Rio de Janeiro Health Regions correlation plot. Source: authors.

Fig. C5 Italy Regions correlation plot. Source: authors.

 Springer Nature 2021 LATEX template

Comparing short-term univariate and multivariate time series forecasting ... 45

Fig. C6 US States correlation plot. Source: authors.

References
[1] Marques, J.A.L., Gois, F.N.B., Xavier-Neto, J., Fong, S.J.: Predictive
Models for Decision Support in the COVID-19 Crisis. Springer, ???
(2021)

[2] Kaur, H., Garg, S., Joshi, H., Ayaz, S., Sharma, S., Bhandari, M.:
A review: Epidemics and pandemics in human history. International
Journal of Pharma Research and Health Sciences 8, 3139–3142 (2020).
https://doi.org/10.21276/ijprhs.2020.02.01

[3] Porta, M.: A Dictionary of Epidemiology. Oxford university press, ???

(2014)

[4] Honigsbaum, M.: Pandemic. The Lancet 373(9679), 1939 (2009)

[5] Yang, Y., Peng, F., Wang, R., Guan, K., Jiang, T., Xu, G., Sun, J.,
Chang, C.: The deadly coronaviruses: The 2003 sars pandemic and the
2020 novel coronavirus epidemic in china. Journal of autoimmunity 109,
102434 (2020)

[6] Chowell, G., Sattenspiel, L., Bansal, S., Viboud, C.: Mathematical mod-
els to characterize early epidemic growth: A review. Physics of life
reviews 18, 66–97 (2016)

[7] Metcalf, C.J.E., Lessler, J.: Opportunities and challenges in modeling

emerging infectious diseases. Science 357(6347), 149–152 (2017)

[8] Smirnova, A., Chowell, G.: A primer on stable parameter estimation

and forecasting in epidemiology by a problem-oriented regularized least
 Springer Nature 2021 LATEX template

46 Comparing short-term univariate and multivariate time series forecasting ...

squares algorithm. Infectious Disease Modelling 2(2), 268–275 (2017)

[9] Hays, J.N.: Epidemics and Pandemics: Their Impacts on Human History.
Abc-clio, ??? (2005)

[10] White, P.: Epidemics and pandemics: their impacts on human history.
Reference Reviews (2006)

[11] Yamey, G., Schäferhoff, M., Aars, O.K., Bloom, B., Carroll, D., Chawla,
M., Dzau, V., Echalar, R., Gill, I.S., Godal, T., et al.: Financing of
international collective action for epidemic and pandemic preparedness.
The Lancet Global Health 5(8), 742–744 (2017)

[12] Chen, Y., Li, Q., Karimian, H., Chen, X., Li, X.: Spatio-temporal dis-
tribution characteristics and influencing factors of covid-19 in china.
Scientific Reports 11(1), 1–12 (2021)

[13] ArunKumar, K., Kalaga, D.V., Kumar, C.M.S., Chilkoor, G., Kawaji,
M., Brenza, T.M.: Forecasting the dynamics of cumulative covid-19 cases
(confirmed, recovered and deaths) for top-16 countries using statisti-
cal machine learning models: Auto-regressive integrated moving average
(arima) and seasonal auto-regressive integrated moving average (sarima).
Applied soft computing 103, 107161 (2021)

[14] Katris, C.: A time series-based statistical approach for outbreak spread
forecasting: Application of covid-19 in greece. Expert Systems with
Applications 166, 114077 (2021)

[15] Benı́tez, D., Montero, G., Rodrı́guez, E., Greiner, D., Oliver, A.,
González, L., Montenegro, R.: A phenomenological epidemic model
based on the spatio-temporal evolution of a gaussian probability density
function. Mathematics 8(11), 2000 (2020)

[16] Hyndman, R.J., Athanasopoulos, G.: Forecasting: Principles and Prac-

tice. OTexts, ??? (2018)

[17] Yule, G.U.: Vii. on a method of investigating periodicities disturbed

series, with special reference to wolfer’s sunspot numbers. Philosophical
Transactions of the Royal Society of London. Series A, Containing Papers
of a Mathematical or Physical Character 226(636-646), 267–298 (1927)

[18] Kermack, W.O., McKendrick, A.G.: A contribution to the mathematical

theory of epidemics. Proceedings of the royal society of london. Series A,
Containing papers of a mathematical and physical character 115(772),
700–721 (1927)

[19] Holt, C.: Forecasting seasonals and trends by exponentially weighted

 Springer Nature 2021 LATEX template

Comparing short-term univariate and multivariate time series forecasting ... 47

averages (onr memorandum no. 52). Carnegie Institute of Technology,

Pittsburgh USA 10 (1957)

[20] Brown, R.G.: Statistical Forecasting for Inventory Control.

McGraw/Hill, ??? (1959)

[21] Winters, P.R.: Forecasting sales by exponentially weighted moving

averages. Management science 6(3), 324–342 (1960)

[22] Box, G., Jenkins, G.: Control. Halden-Day, San Francisco (1970)

[23] Samuel, A.L.: Some studies in machine learning using the game of
checkers. IBM Journal of research and development 3(3), 210–229 (1959)

[24] Kalman, R.E., et al.: Contributions to the theory of optimal control. Bol.
soc. mat. mexicana 5(2), 102–119 (1960)

[25] Chretien, J.-P., George, D., Shaman, J., Chitale, R.A., McKenzie, F.E.:
Influenza forecasting in human populations: a scoping review. PloS one
9(4), 94130 (2014)

[26] Kiang, M.V., Santillana, M., Chen, J.T., Onnela, J.-P., Krieger, N.,
Engø-Monsen, K., Ekapirat, N., Areechokchai, D., Prempree, P., Maude,
R.J., et al.: Incorporating human mobility data improves forecasts of
dengue fever in thailand. Scientific reports 11(1), 1–12 (2021)

[27] Talkhi, N., Fatemi, N.A., Ataei, Z., Nooghabi, M.J.: Modeling and
forecasting number of confirmed and death caused covid-19 in iran:
A comparison of time series forecasting methods. Biomedical Signal
Processing and Control 66, 102494 (2021)

[28] Khan, F., Saeed, A., Ali, S.: Modelling and forecasting of new cases,
deaths and recover cases of covid-19 by using vector autoregressive model
in pakistan. Chaos, Solitons & Fractals 140, 110189 (2020)

[29] Bomfim, R., Pei, S., Shaman, J., Yamana, T., Makse, H.A., Andrade Jr,
J.S., Lima Neto, A.S., Furtado, V.: Predicting dengue outbreaks at neigh-
bourhood level using human mobility in urban areas. Journal of the
Royal Society Interface 17(171), 20200691 (2020)

[30] Liang, X., Xu, Q., Guan, R., Zhao, Y.: Forecasting tuberculosis incidence
in china using baidu index: A comparative study. In: Proceedings of the
4th International Conference on Medical and Health Informatics, pp.
22–29 (2020)

[31] Ramos, A.C.V., Gomes, D., Santos Neto, M., Berra, T.Z., de Assis,
I.S., Yamamura, M., Crispim, J.d.A., Martoreli Junior, J.F., Bruce,
 Springer Nature 2021 LATEX template

48 Comparing short-term univariate and multivariate time series forecasting ...

A.T.I., Dos Santos, F.L., et al.: Trends and forecasts of leprosy for a
hyperendemic city from brazil’s northeast: Evidence from an eleven-year
time-series analysis. PloS one 15(8), 0237165 (2020)

[32] Zhang, C., Fu, X., Zhang, Y., Nie, C., Li, L., Cao, H., Wang, J., Wang, B.,
Yi, S., Ye, Z.: Epidemiological and time series analysis of haemorrhagic
fever with renal syndrome from 2004 to 2017 in shandong province, china.
Scientific reports 9(1), 1–9 (2019)

[33] Wang, Y., Xu, C., Zhang, S., Yang, L., Wang, Z., Zhu, Y., Yuan, J.:
Development and evaluation of a deep learning approach for modeling
seasonality and trends in hand-foot-mouth disease incidence in mainland
china. Scientific reports 9(1), 1–15 (2019)

[34] Li, K., Liu, M., Feng, Y., Ning, C., Ou, W., Sun, J., Wei, W., Liang, H.,
Shao, Y.: Using baidu search engine to monitor aids epidemics inform for
targeted intervention of hiv/aids in china. Scientific reports 9(1), 1–12
(2019)

[35] Choi, S.B., Kim, J., Ahn, I.: Forecasting type-specific seasonal influenza
after 26 weeks in the united states using influenza activities in other
countries. PLoS One 14(11), 0220423 (2019)

[36] Chakraborty, T., Chattopadhyay, S., Ghosh, I.: Forecasting dengue epi-
demics using a hybrid methodology. Physica A: Statistical Mechanics
and its Applications 527, 121266 (2019)

[37] Chumachenko, D., Turiy, A., Chukhray, A.: Application of statistical

simulation for measles epidemic process forecasting. In: 2019 IEEE 2nd
Ukraine Conference on Electrical and Computer Engineering (UKR-
CON), pp. 1086–1090 (2019). IEEE

[38] Haddawy, P., Yin, M.S., Wisanrakkit, T., Limsupavanich, R., Promrat,
P., Lawpoolsri, S., Sa-angchai, P.: Complexity-based spatial hierarchi-
cal clustering for malaria prediction. Journal of Healthcare Informatics
Research 2(4), 423–447 (2018)

[39] Wu, H., Wang, X., Xue, M., Wu, C., Lu, Q., Ding, Z., Zhai, Y., Lin, J.:
Spatial-temporal characteristics and the epidemiology of haemorrhagic
fever with renal syndrome from 2007 to 2016 in zhejiang province, china.
Scientific reports 8(1), 1–14 (2018)

[40] Wu, Y., Yang, Y., Nishiura, H., Saitoh, M.: Deep learning for epidemi-
ological predictions. In: The 41st International ACM SIGIR Conference
on Research & Development in Information Retrieval, pp. 1085–1088
(2018)
 Springer Nature 2021 LATEX template

Comparing short-term univariate and multivariate time series forecasting ... 49

[41] Zhao, Y., Ge, L., Zhou, Y., Sun, Z., Zheng, E., Wang, X., Huang, Y.,
Cheng, H.: A new seasonal difference space-time autoregressive inte-
grated moving average (sd-starima) model and spatiotemporal trend
prediction analysis for hemorrhagic fever with renal syndrome (hfrs).
PloS one 13(11), 0207518 (2018)

[42] Jerónimo-Martı́nez, L.E., Menéndez-Mora, R.E., Bolı́var, H.: Forecasting

acute respiratory infection cases in southern bogota: Ears vs. arima and
sarima. In: 2017 Congreso Internacional de Innovacion Y Tendencias en
Ingenieria (CONIITI), pp. 1–6 (2017). IEEE

[43] Ray, E.L., Sakrejda, K., Lauer, S.A., Johansson, M.A., Reich, N.G.:
Infectious disease prediction with kernel conditional density estimation.
Statistics in medicine 36(30), 4908–4929 (2017)

[44] Anggraeni, W., Aristiani, L.: Using google trend data in forecast-
ing number of dengue fever cases with arimax method case study:
Surabaya, indonesia. In: 2016 International Conference on Information
& Communication Technology and Systems (ICTS), pp. 114–118 (2016).
IEEE

[45] Ke, G., Hu, Y., Huang, X., Peng, X., Lei, M., Huang, C., Gu, L., Xian,
P., Yang, D.: Epidemiological analysis of hemorrhagic fever with renal
syndrome in china with the seasonal-trend decomposition method and
the exponential smoothing model. Scientific reports 6(1), 1–7 (2016)

[46] Li, S., Cao, W., Ren, H., Lu, L., Zhuang, D., Liu, Q.: Time series anal-
ysis of hemorrhagic fever with renal syndrome: A case study in jiaonan
county, china. Plos one 11(10), 0163771 (2016)

[47] Johansson, M.A., Reich, N.G., Hota, A., Brownstein, J.S., Santillana, M.:
Evaluating the performance of infectious disease forecasts: A comparison
of climate-driven and seasonal dengue forecasts for mexico. Scientific
reports 6(1), 1–11 (2016)

[48] Pradhan, A., Anasuya, A., Pradhan, M.M., Ak, K., Kar, P., Sahoo, K.C.,
Panigrahi, P., Dutta, A.: Trends in malaria in odisha, india—an analysis
of the 2003–2013 time-series data from the national vector borne disease
control program. PloS one 11(2), 0149126 (2016)

[49] Wu, W., Guo, J., An, S., Guan, P., Ren, Y., Xia, L., Zhou, B.: Compar-
ison of two hybrid models for forecasting the incidence of hemorrhagic
fever with renal syndrome in jiangsu province, china. PLoS One 10(8),
0135492 (2015)

[50] Mekparyup, J., Saithanu, K.: Forecasting the dengue hemorrhagic fever
cases using seasonal arima model in chonburi, thailand. Global J Pure
 Springer Nature 2021 LATEX template

50 Comparing short-term univariate and multivariate time series forecasting ...

Appl Math 11, 401–407 (2015)

[51] Kane, M.J., Price, N., Scotch, M., Rabinowitz, P.: Comparison of arima
and random forest time series models for prediction of avian influenza
h5n1 outbreaks. BMC bioinformatics 15(1), 1–9 (2014)

[52] Feng, H., Duan, G., Zhang, R., Zhang, W.: Time series analysis of
hand-foot-mouth disease hospitalization in zhengzhou: establishment of
forecasting models using climate variables as predictors. PloS one 9(1),
87916 (2014)

[53] Soebiyanto, R.P., Adimi, F., Kiang, R.K.: Modeling and predicting
seasonal influenza transmission in warm regions using climatological
parameters. PloS one 5(3), 9450 (2010)

[54] Shen, Y., Jiang, C., Dun, Z.: Analysis and prediction of epidemiological
trend of scarlet fever from 1957 to 2004 in the downtown area of beijing.
In: International Workshop on Biosurveillance and Biosecurity, pp. 164–
168 (2008). Springer

[55] Medina, D.C., Findley, S.E., Guindo, B., Doumbia, S.: Forecasting non-
stationary diarrhea, acute respiratory infection, and malaria time-series
in niono, mali. PLoS One 2(11), 1181 (2007)

[56] Burkom, H.S., Murphy, S.P., Shmueli, G.: Automated time series fore-
casting for biosurveillance. Statistics in medicine 26(22), 4202–4218
(2007)

[57] Nobre, F.F., Monteiro, A.B.S., Telles, P.R., Williamson, G.D.: Dynamic
linear model and sarima: a comparison of their forecasting performance
in epidemiology. Statistics in medicine 20(20), 3051–3069 (2001)

[58] Krause, A.L., Kurowski, L., Yawar, K., Van Gorder, R.A.: Stochastic
epidemic metapopulation models on networks: Sis dynamics and control
strategies. Journal of theoretical biology 449, 35–52 (2018)

[59] Gerardi, D., Monteiro, L.: System identification and prediction of dengue
fever incidence in rio de janeiro. Mathematical Problems in Engineering
2011 (2011)

[60] Paul, A., Reja, S., Kundu, S., Bhattacharya, S.: Covid-19 pandemic
models revisited with a new proposal: Plenty of epidemiological mod-
els outcast the simple population dynamics solution. Chaos, Solitons &
Fractals 144, 110697 (2021)

[61] Wang, P., Zheng, X., Li, J., Zhu, B.: Prediction of epidemic trends
in covid-19 with logistic model and machine learning technics. Chaos,
 Springer Nature 2021 LATEX template

Comparing short-term univariate and multivariate time series forecasting ... 51

Solitons & Fractals 139, 110058 (2020)

[62] Smirnova, A., Sirb, B., Chowell, G.: On stable parameter estimation
and forecasting in epidemiology by the levenberg–marquardt algorithm
with broyden’s rank-one updates for the jacobian operator. Bulletin of
mathematical biology 81(10), 4210–4232 (2019)

[63] Eilertson, K.E., Fricks, J., Ferrari, M.J.: Estimation and prediction
for a mechanistic model of measles transmission using particle filter-
ing and maximum likelihood estimation. Statistics in medicine 38(21),
4146–4158 (2019)

[64] Suparit, P., Wiratsudakul, A., Modchang, C.: A mathematical model for
zika virus transmission dynamics with a time-dependent mosquito biting
rate. Theoretical Biology and Medical Modelling 15(1), 1–11 (2018)

[65] Basile, L., Oviedo de la Fuente, M., Torner, N., Martı́nez, A., Jané, M.:
Real-time predictive seasonal influenza model in catalonia, spain. PloS
one 13(3), 0193651 (2018)

[66] Li, X., Doroshenko, A., Osgood, N.D.: Applying particle filtering in both
aggregated and age-structured population compartmental models of pre-
vaccination measles. PloS one 13(11), 0206529 (2018)

[67] Valeri, L., Patterson-Lomba, O., Gurmu, Y., Ablorh, A., Bobb, J.,
Townes, F.W., Harling, G.: Predicting subnational ebola virus disease
epidemic dynamics from sociodemographic indicators. PloS one 11(10),
0163544 (2016)

[68] Yang, W., Karspeck, A., Shaman, J.: Comparison of filtering methods
for the modeling and retrospective forecasting of influenza epidemics.
PLoS computational biology 10(4), 1003583 (2014)

[69] Wang, X., Panchanathan, S., Chowell, G.: A data-driven mathematical

model of ca-mrsa transmission among age groups: evaluating the effect
of control interventions. PLoS computational biology 9(11), 1003328
(2013)

[70] Towers, S., Chowell, G.: Impact of weekday social contact patterns on the
modeling of influenza transmission, and determination of the influenza
latent period. Journal of theoretical biology 312, 87–95 (2012)

[71] Aguiar, M., Ballesteros, S., Kooi, B.W., Stollenwerk, N.: The role of
seasonality and import in a minimalistic multi-strain dengue model cap-
turing differences between primary and secondary infections: complex
dynamics and its implications for data analysis. Journal of theoretical
biology 289, 181–196 (2011)
 Springer Nature 2021 LATEX template

52 Comparing short-term univariate and multivariate time series forecasting ...

[72] Laneri, K., Bhadra, A., Ionides, E.L., Bouma, M., Dhiman, R.C., Yadav,
R.S., Pascual, M.: Forcing versus feedback: epidemic malaria and mon-
soon rains in northwest india. PLoS computational biology 6(9), 1000898
(2010)

[73] Santos, L., Costa, M., Pinho, S.T.R.d., Andrade, R.F.S., Barreto, F.R.,
Teixeira, M., Barreto, M.L.: Periodic forcing in a three-level cellular
automata model for a vector-transmitted disease. Physical Review E
80(1), 016102 (2009)

[74] Finkenstädt, B., Morton, A., Rand, D.: Modelling antigenic drift in
weekly flu incidence. Statistics in medicine 24(22), 3447–3461 (2005)

[75] Gamerman, D., Migon, H.S.: Forecasting the number of aids cases
in brazil. Journal of the Royal Statistical Society: Series D (The
Statistician) 40(4), 427–442 (1991)

[76] Koller, D., Friedman, N.: Probabilistic Graphical Models: Principles and
Techniques. MIT press, ??? (2009)

[77] Haykin, S.: Kalman Filtering and Neural Networks vol. 47. John Wiley
& Sons, ??? (2004)

[78] Han, T., Gois, F.N.B., Oliveira, R., Prates, L.R., de Almeida Porto,
M.M.: Modeling the progression of covid-19 deaths using kalman filter
and automl. Soft Computing, 1–16 (2021)

[79] Nunes, B., Natário, I., Lucı́lia Carvalho, M.: Nowcasting influenza epi-
demics using non-homogeneous hidden markov models. Statistics in
Medicine 32(15), 2643–2660 (2013)

[80] Mode, C.J., Fife, D., Troy, S.M.: Stochastic methods for short term
projections of symptomatic hiv disease. Statistics in medicine 10(9),
1427–1440 (1991)

[81] Ongsulee, P.: Artificial intelligence, machine learning and deep learn-
ing. In: 2017 15th International Conference on ICT and Knowledge
Engineering (ICT&KE), pp. 1–6 (2017). IEEE

[82] Kohavi, R.: Glossary of terms. Special issue on applications of machine

learning and the knowledge discovery process 30(271), 127–132 (1998)

[83] Alzubi, J., Nayyar, A., Kumar, A.: Machine learning from theory to
algorithms: an overview. In: Journal of Physics: Conference Series, vol.
1142, p. 012012 (2018). IOP Publishing

[84] Ribeiro, M.H.D.M., Mariani, V.C., dos Santos Coelho, L.: Multi-step
 Springer Nature 2021 LATEX template

Comparing short-term univariate and multivariate time series forecasting ... 53

ahead meningitis case forecasting based on decomposition and multi-

objective optimization methods. Journal of Biomedical Informatics 111,
103575 (2020)

[85] Deng, S., Wang, S., Rangwala, H., Wang, L., Ning, Y.: Cola-gnn:
Cross-location attention based graph neural networks for long-term ili
prediction. In: Proceedings of the 29th ACM International Conference
on Information & Knowledge Management, pp. 245–254 (2020)

[86] Stolerman, L.M., Maia, P.D., Kutz, J.N.: Forecasting dengue fever in
brazil: An assessment of climate conditions. PloS one 14(8), 0220106
(2019)

[87] Caicedo-Torres, W., Montes-Grajales, D., Miranda-Castro, W., Fennix-

Agudelo, M., Agudelo-Herrera, N.: Kernel-based machine learning mod-
els for the prediction of dengue and chikungunya morbidity in colombia.
In: Colombian Conference on Computing, pp. 472–484 (2017). Springer

[88] Nguyen, H.L., Duong, T.H., Nguyen, C.P., Nguyen, D.C., Chiem,
T.P., Nguyen, M.H., Nguyen, T.N.M., Nguyen, H.V.: Specific k-mean
clustering-based perceptron for dengue prediction. International Jour-
nal of Intelligent Information and Database Systems 10(3-4), 269–288
(2017)

[89] Chau, N.H., Ngoc Anh, L.T.: Using local weather and geographi-
cal information to predict cholera outbreaks in hanoi, vietnam. In:
Advanced Computational Methods for Knowledge Engineering, pp.
195–212. Springer, ??? (2016)

[90] Peng, L.-Z., Yi, L.-X., Hua, S.-Y.: A new epidemic disease predicting
method. In: 2008 International Conference on Intelligent Computation
Technology and Automation (ICICTA), vol. 1, pp. 550–553 (2008). IEEE

[91] Fekedulegn, D., Mac Siúrtáin, M.P., Colbert, J.J.: Parameter estimation
of nonlinear models in forestry. Silva Fennica 33(4), 327–336 (1999)

[92] Kaps, M., Herring, W., Lamberson, W.: Genetic and environmental
parameters for traits derived from the brody growth curve and their rela-
tionships with weaning weight in angus cattle. Journal of Animal Science
78(6), 1436–1442 (2000)

[93] Tsoularis, A., Wallace, J.: Analysis of logistic growth models. Mathe-
matical biosciences 179(1), 21–55 (2002)

[94] Khamis, A.: Nonlinear growth models for modeling oil palm yield growth.
J. of mathematics and statistics 1(3), 225–233 (2005)
 Springer Nature 2021 LATEX template

54 Comparing short-term univariate and multivariate time series forecasting ...

[95] Assad, D.B.N.: Short-term-forecasting. GitHub (2022)

[96] Krispin, R.: Covid19italy: The 2019 Novel Coronavirus COVID-19 (2019-
nCoV) Italy Dataset. (2021). R package version 0.3.1. https://CRAN.
R-project.org/package=covid19italy

[97] Dobbyn, A.: Covid19us: Cases of COVID-19 in the United States.

(2020). R package version 0.1.7. https://CRAN.R-project.org/package=
covid19us

[98] Kotu, V., Deshpande, B.: Time series forecasting. Data Science; Elsevier:
Amsterdam, The Netherlands, 395–445 (2019)

[99] De Livera, A.M., Hyndman, R.J., Snyder, R.D.: Forecasting time series
with complex seasonal patterns using exponential smoothing. Journal of
the American statistical association 106(496), 1513–1527 (2011)

[100] Tabataba, F.S., Lewis, B., Hosseinipour, M., Tabataba, F.S., Venkatra-
manan, S., Chen, J., Higdon, D., Marathe, M.: Epidemic forecasting
framework combining agent-based models and smart beam particle fil-
tering. In: 2017 IEEE International Conference on Data Mining (ICDM),
pp. 1099–1104 (2017). IEEE

[101] Chuang, T.-W., Chaves, L.F., Chen, P.-J.: Effects of local and regional
climatic fluctuations on dengue outbreaks in southern taiwan. PLoS One
12(6), 0178698 (2017)

[102] Guo, P., Zhang, J., Wang, L., Yang, S., Luo, G., Deng, C., Wen, Y.,
Zhang, Q.: Monitoring seasonal influenza epidemics by using internet
search data with an ensemble penalized regression model. Scientific
reports 7(1), 1–11 (2017)