Covid-19 in Germany: A Case Study

Abrar Ahmed

University of Trier
[email protected]

February 23, 2021

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 1 / 44

Overview
1 Introduction
World-wide Pandemic!
Covid-19 in Germany
Aim of the study
2 Literature Review
3 Data Description
4 Direction of the Project
Exploratory Data Analysis and Visualization
Prediction
Panel Data Regression
Fixed Effects Model
Random effects model
SEIR Model
Artificial Neural Network
Recurrent Neural Network
5 Analysis and Results (so far...)
6 Problems Faced...
7 Work Still to be done
Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 2 / 44
 Section 1

Introduction

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 3 / 44

credit: https:
//www.weser-kurier.de/deutschland-welt/deutschland-welt-politik_artikel,
-liveticker-verteilung-der-ffp2masken-im-land-bremen-abgeschlossen-_arid,
1899754.html
Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 4 / 44
World-wide Pandemic!

The biggest talking point of the last one year.

A contagious disease caused by severe acute respiratory syndrome
coronavirus 2 (SARS-CoV-2).
Fever, cough, fatigue, breathing difficulties, and loss of smell and taste.
Spreads mainly when an infected person is in close contact with
another person.
Virus gets into their mouth, nose or eyes.
First identified in December 2019 in Wuhan, China.
Pandemic declared in March 2020
Over 111 million recorded cases as of late.
The world is at a standstill.

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 5 / 44

Covid-19 in Germany

The first case was confirmed near München, Bavaria, 27 January,

2020.
First 2 cases of death reported in Essen and Heinsberg, on 9 March,
2020.
As of late, over 2 million confirmed cases, and over 65 thousand
deaths.

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 6 / 44

Aim of the study

The scenario of the pandemic is ever-changing.

New strands of the virus.
Mobility.
Lockdown.
Weather conditions.
Research Question:
How is the current situation of the Pandemic in Germany?
How well we can predict the condition of the Pandemic?

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 7 / 44

 Section 2

Literature Review

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 8 / 44

Literature Review
Quantified the effects of mobility habits in the spread of the Coronavirus in Italy
through a multiple linear regression model. Estimation results showed that mobility
habits represent one of the variables that explains the number of COVID-19
infections jointly with the number of tests/day and some environmental variables.
[Cartenı̀, Di Francesco, and Martino, 2020].

First study to develop both time series and panel data models to construct
conceptual tools that examine the nexus between death from COVID-19 and
confirmed cases.[Sarkodie and Owusu, 2020]

This study considered a gravity model that expressed time to first case as a function
of multiple socio-economic factors. [Vigdorovits, 2020]

SEIR model and Regression model have been used for predictions based on the data
collected from John Hopkins University repository during a certain time period.
[Pandey et al., 2020] Data-driven estimation methods like long short-term memory

(LSTM) have been used and curve fitting for prediction of the number of COVID-19
cases in India 30 days ahead [Tomar and Gupta, 2020]
Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 9 / 44
 Section 3

Data Description

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 10 / 44

Data Description

Panel Data:
The term “panel data” refers to the pooling of observations on a
cross-section of households, countries, firms, etc. over several time
periods. This can be achieved by surveying a number of households or
individuals and following them over time.[Baltagi, 2008]
Data ranging from 01.03.2020 till 31.10.2020
Variables
Date
Landkreis
Cases per day
Deaths
7 day Incidence per 100,000
Average Temperature per month
Total number of Mobility flows per day
Precipitation
Hours of sunlight

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 11 / 44

Data Source

Covid data taken from the RKI data platform 1

Weather data collected from Corona-Daten-Plattform 2

Mobility Data from Teralytics 3

1
https://npgeo-corona-npgeo-
de.hub.arcgis.com/datasets/dd4580c810204019a7b8eb3e0b329dd60
2
https://www.corona-datenplattform.de/
3
https://www.teralytics.net/
Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 12 / 44
 Section 4

Direction of the Project

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 13 / 44

Direction of the Project

The project has two parts:

Exploratory Data Analysis and Visualization
Prediction

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 14 / 44

Exploratory Data Analysis and Visualization

Basic Visualizations of the number of cases, deaths etc.

Visualizations over time
Across several Bundeslands and Landkreis

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 15 / 44

Prediction

Panel Data Regression

SEIR Model
Neural Network

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 16 / 44

Panel Data Regression

Panel Data have both cross-sectional and time components

Balanced and Unbalanced Panel Data
The standard static model with i = 1, ..., N, t = 1, ..., T is

yit = β0 + x0it β + it

xit = K -dimensional vector of explanatory variables, without a const term.

β0 = intercept, independent of i and t.
β,a(K × 1) vector, the slopes, is independent of i and t.
it = the error, varies over i and t.

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 17 / 44

Problems:
Consistency/Exogenity
Autocorrelation
Solution:
iid 0, σu2


it = αi + uit with uit

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 18 / 44

Fixed effects model4 , FE: αi are individual intercepts (fixed for given N).

yit = αi + x0it β + uit

No overall intercept is (usually) included in the model

Under FE, consistency does not require, that the individual intercepts
(whose coefficients are the αi ’s) and uit are uncorrelated. Only
E (xit uit ) = 0 must hold.
There are N − 1 additional parameters for capturing the individual
heteroscedasticity.

4
http://statmath.wu.ac.at/ hauser/LVs/FinEtricsQF/FEtricsC hp5.pdf
Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 19 / 44
Random effects model5 , RE:

αi ∼ iid 0, σα2

yit = β0 + x0it β + αi + uit , uit ∼ iid 0, σu2

The αi ’s are rvs with the same variance. The value αi is specific for
individual i. The α’s of different individualss are independent, have a mean
of zero, and their distribution is assumed to be not too far away from
normality. The overall mean is captured in β0 . αi is time invariant and
homoscedastic across individuals.
There is only one additional parameter σα2 . Only αi contributes to
Corr (i,s , i,t ). αi determines both i,s and i,t

5
http://statmath.wu.ac.at/ hauser/LVs/FinEtricsQF/FEtricsC hp5.pdf
Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 20 / 44
SEIR Model

Susceptible-Exposed-Infectious-Recovered (SEIR) Model is an

epidemiological model
WHO used SEIR models to characterize and forecast the early stages
of the COVID-19 outbreak in Wuhan.

Figure: SEIR model

source: www.triplebyte.com/blog/modeling-infectious-diseases

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 21 / 44

 S = fraction of susceptible individuals (those able to contract the
disease)
E = fraction of exposed individuals (those who have been infected
but are not yet infectious),
I = fraction of infected individuals (those capable of transmitting the
disease),
R = fraction of recovered individuals (those who have become
immune).
(Values are normalized)

S + E + I + R = 1.

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 22 / 44

Assumptions
Equal birth and death rates µ
1/α = mean latent period for the disease
1/γ = mean infectious period recovered individuals are permanently
immune
= contact rate β
We get the model:
dS
= µ − β(t)SI − µS
dt
dE
= β(t)SI − (µ + α)E
dt
dI
= αE − (µ + γ)I
dt
We get the value R0 as:
β0 α
R0 = ,
(µ + α)(µ + γ)

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 23 / 44

   
Number of Probability of  
Duration of
R0 =  contacts   transmission 
infection
per unit time per contact
 
Probabililty of
× surviving 
exposed stage

If R0 < 1, the disease-free equilibrium point is globally asymptotically

stable and there is no endemic equilibrium point (the disease dies out).
If R0 > 1, the disease-free equilibrium point is unstable and a globally
asymptotically stable endemic equilibrium point exists.

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 24 / 44

Artificial Neural Network

Computational model that is inspired by the way biological neural

networks in the human brain process information
The node receives input from some other nodes, or from an external
source and computes an output
Weight assigned to input on the basis of importance
Node applies a function to the weighted sum of its inputs
If the final sum is above a certain threshold, the neuron can fire
(activation function)

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 25 / 44

 Figure: Artificial Neural Network

credit: https://www.extremetech.com/extreme/
215170-artificial-neural-networks-are-changing-the-world-what-are-they
Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 26 / 44
Recurrent Neural Network
Feedforward Neural Network
Connections between units do not form a cycle
Information moves in only one direction
No cycles or loop
Recurrent Neural Network:
Generalization of Feedforward Neural Network
Recurrent in nature
Considers current input and output for decision making

Figure: Recurrent Neural Network

https://towardsdatascience.com/understanding-rnn-and-lstm-f7cdf6dfc14e
Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 27 / 44
Backpropagation
Algorithm for training Feedforward Neural Network
Backpropagation strictly refers only to the algorithm for computing
the gradient

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 28 / 44

LSTM
Long-Short-Term-Memory
Modified version of RNN
Trains model by back-propagation

Figure: LSTM

https://towardsdatascience.com/understanding-rnn-and-lstm-f7cdf6dfc14e
Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 29 / 44
 Section 5

Analysis and Results (so far...)

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 30 / 44

Analysis and Results (so far...)

Figure: Cases all over the world

credit: www.google.com
Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 31 / 44
Germany with respect to other countries

Total cases = 112 mil (approx)

Recovered = 63 mil (approx)
Deaths = 2.48 mil (approx)
Germany ranks 10th with 2.4 mil cases and 68k (approx) deaths
3883 new cases
415 new deaths (As of 22.02.2021)

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 32 / 44

Cases and Deaths in Germany
Figure: Daily Cases in Germany

Figure: Daily Deaths in Germany

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 33 / 44

Cases per Landkreis

Figure: Cases in every Landkreis

credit: https://experience.arcgis.com/experience/
478220a4c454480e823b17327b2bf1d4/page/page_1/
Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 34 / 44
 Table: Corona Inceidence per Landkries
7-day Incidence 7-day-count Landkreis
355.3 256 LK Tirschenreuth (Bavaria)
306.9 223 LK Wunsiedel i.Fichtelgebirge (Bavaria)
252.7 108 SK Weiden i.d.OPf. (Bavaria)
216.9 271 LK Schmalkalden-Meiningen (Thuringia)
199.4 189 LK Hof (Bavaria)
196.3 351 LK Burgenlandkreis (Sachsen-Anhalt)
. . .
. . .
. . .
30.5 34 SK Trier (Rhineland Pfalz)
. . .
. . .
. . .

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 35 / 44

S-E-I-R Model Analysis

Figure: SEIR Model Prediction

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 36 / 44

S-E-I-R Model Analysis after intervention

Figure: SEIR Model Prediction

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 37 / 44

 Section 6

Problems Faced...

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 38 / 44

Problems faced

Huge Volume of Data

Huge burden on system
Multiple Data Sources
Handling of Panel Data much more difficult in Python

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 39 / 44

 Section 7

Work Still to be done

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 40 / 44

Work Still to be done

Perform the Panel Data Regression Analysis

Perform the Neural Network Analysis
Incorporate fine Tuning for Holidays

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 41 / 44

 Thank You!

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 42 / 44

References I

Baltagi, Badi (2008). Econometric analysis of panel data. John Wiley &
Sons.
Cartenı̀, Armando, Luigi Di Francesco, and Maria Martino (2020). “How
mobility habits influenced the spread of the COVID-19 pandemic:
Results from the Italian case study”. In: Science of the Total
Environment 741, p. 140489.
Pandey, Gaurav et al. (2020). “SEIR and Regression Model based
COVID-19 outbreak predictions in India”. In: arXiv preprint
arXiv:2004.00958.
Sarkodie, Samuel Asumadu and Phebe Asantewaa Owusu (2020).
“Investigating the cases of novel coronavirus disease (COVID-19) in
China using dynamic statistical techniques”. In: Heliyon 6.4, e03747.
Tomar, Anuradha and Neeraj Gupta (2020). “Prediction for the spread of
COVID-19 in India and effectiveness of preventive measures”. In: Science
of The Total Environment 728, p. 138762.

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 43 / 44

References II

Vigdorovits, Alon (2020). “Buying time: an ecological survival analysis of

COVID-19 spread based on the gravity model”. In: medRxiv.

Abrar Ahmed (University of Trier) Covid-19 in Germany February 23, 2021 44 / 44