COVID-19 Data Analysis - Predicting the Evolution in Brazil with ARIMA & Prophet

I used the data generated from previous ETL code and available on CSSEGISandData Github - https://raw.githubusercontent.com/CSSEGISandData/COVID-19/master/csse_covid_19_data/csse_covid_19_time_series/time_series_covid19_deaths_global.csv

# libraries
#!pip install plotly
import pandas as pd
import pandera as pa
import numpy as np
from datetime import datetime
import plotly.express as px
import plotly.graph_objects as go

Importing & Filtering Dataset

# Importing data
# Confirmed
url_conf = 'https://raw.githubusercontent.com/CSSEGISandData/COVID-19/master/csse_covid_19_data/csse_covid_19_time_series/time_series_covid19_confirmed_global.csv'
df_conf = pd.read_csv(url_conf)
# Deaths
url_deaths = 'https://raw.githubusercontent.com/CSSEGISandData/COVID-19/master/csse_covid_19_data/csse_covid_19_time_series/time_series_covid19_deaths_global.csv'
df_deaths = pd.read_csv(url_deaths)

# Changing shape of the df - passing dates to lines
data_conf = pd.melt(df_conf, id_vars=['Province/State','Country/Region', 'Lat', 'Long'], 
var_name='Date', value_name='Confirmed')
data_conf.columns = ['Province', 'Country', 'Lat', 'Long', 'Dates', 'Confirmed']

# Changing shape of the df - passing dates to lines
data_deaths = pd.melt(df_deaths, id_vars=['Province/State','Country/Region', 'Lat', 'Long'], 
var_name='Date', value_name='Deaths')
data_deaths.columns = ['Province', 'Country', 'Lat', 'Long', 'Dates', 'Deaths']

# Removing unuseful cols
data_conf.drop(['Province', 'Lat', 'Long'], axis = 1, inplace = True)
data_deaths.drop(['Province', 'Lat', 'Long'], axis = 1, inplace = True)

# Checking data schema
data_conf['Dates'] = pd.to_datetime(data_conf['Dates'], format="%m/%d/%y")

schema = pa.DataFrameSchema(
    columns = {
        "Country": pa.Column(pa.String),
        "Dates": pa.Column(pa.DateTime),
        "Confirmed": pa.Column(pa.Int)
    }
)

schema.validate(data_conf)

data_deaths['Dates'] = pd.to_datetime(data_deaths['Dates'], format="%m/%d/%y")

schema = pa.DataFrameSchema(
    columns = {
        "Country": pa.Column(pa.String),
        "Dates": pa.Column(pa.DateTime),
        "Deaths": pa.Column(pa.Int)
    }
)

schema.validate(data_deaths)

	Country	Dates	Deaths
0	Afghanistan	2020-01-22	0
1	Albania	2020-01-22	0
2	Algeria	2020-01-22	0
3	Andorra	2020-01-22	0
4	Angola	2020-01-22	0
...	...	...	...
159862	Vietnam	2021-08-16	6141
159863	West Bank and Gaza	2021-08-16	3625
159864	Yemen	2021-08-16	1405
159865	Zambia	2021-08-16	3538
159866	Zimbabwe	2021-08-16	4109

159867 rows × 3 columns

# Binding dfs
data_conf['Deaths'] = data_deaths['Deaths']
data_conf.head()

	Country	Dates	Confirmed	Deaths
0	Afghanistan	2020-01-22	0	0
1	Albania	2020-01-22	0	0
2	Algeria	2020-01-22	0	0
3	Andorra	2020-01-22	0	0
4	Angola	2020-01-22	0	0

# Selecting Brazil
filter1 = (data_conf.Country=='Brazil') & (data_conf.Deaths>0) & (data_conf.Confirmed>0)

data = data_conf.loc[filter1] 
data.head()

	Country	Dates	Confirmed	Deaths
15375	Brazil	2020-03-17	321	1
15654	Brazil	2020-03-18	372	3
15933	Brazil	2020-03-19	621	6
16212	Brazil	2020-03-20	793	11
16491	Brazil	2020-03-21	1021	15

Confirmed Cases

# Evolution of confirmed cases
px.line(data, 'Dates', 'Confirmed', title='Confirmed Cases in Brazil')

New Cases by Day

data['newcases']=list(map(
    lambda x: 0 if (x==0) else data['Confirmed'].iloc[x] - data['Confirmed'].iloc[x-1],
    np.arange(data.shape[0])
))

<ipython-input-8-32dc8658fb5c>:1: SettingWithCopyWarning:


A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy

px.line(data, x='Dates', y='newcases', title='New Cases by Day')

Deaths

fig = go.Figure()

fig.add_trace(
    go.Scatter(x=data.Dates, y=data.Deaths, name='Deaths',
              mode='lines+markers', line={'color':'red'})
)

# layout
fig.update_layout(title='Deaths by COVID-19 in Brazil')

fig.show()

Growing Rate

#growing_rate = (present/past)**(1/n)-1

def growing_rate(data, var, startdate=None, enddate=None):
    # if startdate = None
    if startdate == None:
        startdate = data.Dates.loc[data[var]>0].min()
    else:
        startdate = pd.to_datetime(startdate)
    if enddate == None:
        enddate = data.Dates.iloc[-1]
    else:
        enddate = pd.to_datetime(enddate)
    
    # define present and past values
    past = data.loc[data.Dates == startdate, var].values[0]
    present = data.loc[data.Dates == enddate, var].values[0]
    
    # define number of time points will be evaluated
    n = (enddate - startdate).days
    
    # calculate ratio
    ratio = (present/past)**(1/n)-1
    
    return ratio*100

Mean of Growth Ratio of Covid in Brazil

growing_rate(data, 'Confirmed')

2.1620254335131817

def growing_ratio_day(data, var, startdate=None):
    # if startdate = None
    if startdate == None:
        startdate = data.Dates.loc[data[var]>0].min()
    else:
        startdate = pd.to_datetime(startdate)
    
    enddate = data.Dates.max()
    
    # number of time points evaluated
    n = (enddate-startdate).days
    
    # ratio by day
    ratios = list(map(
        lambda x: (data[var].iloc[x] - data[var].iloc[x-1])/data[var].iloc[x-1],
        range(1, n+1)
    ))
    return np.array(ratios)*100

day_ratio = growing_ratio_day(data, 'Confirmed')

#day_ratio

firstday = data.Dates.loc[data.Confirmed>0].min()

px.line(x=pd.date_range(firstday, data.Dates.max())[1:],
       y=day_ratio, title='Growth Ratio of Confirmed Cases in Brazil',
       labels={'y':'Growth Ratio', 'x':'Date'})

Prediction of Confirmed Cases

from statsmodels.tsa.seasonal import seasonal_decompose
import matplotlib.pyplot as plt

confirmed = data.Confirmed
confirmed.index = data.Dates
res = seasonal_decompose(confirmed)

fig, (ax1, ax2, ax3, ax4) = plt.subplots(4, 1, figsize=(18,8))

ax1.plot(res.observed)
ax2.plot(res.trend)
ax3.plot(res.seasonal)
ax4.plot(confirmed.index, res.resid)
ax4.axhline(0, linestyle='dashed', c='black')
plt.show()

ARIMA Model

#!pip install pyramid-arima
from pmdarima.arima import auto_arima

model =  auto_arima(confirmed)

fig = go.Figure(go.Scatter(
    x=confirmed.index, y=confirmed, name='Observed'
))

fig.add_trace(go.Scatter(
    x=confirmed.index, y=model.predict_in_sample(), name='Predicted'
))

fig.add_trace(go.Scatter(
    x=pd.date_range('2021-08-17', '2021-09-17'), y=model.predict(31), name='Forecast'
))

fig.update_layout(title='Predictions of Confirmed Cases in Brazil to Next 30 Days')
fig.show()

Growth Model with fbprophet

#!pip install pystan==2.19.1.1 prophet

from prophet import Prophet

train = confirmed.reset_index()[:-5]
test = confirmed.reset_index()[-5:]

# renaming columns
train.rename(columns={'Dates': 'ds', 'Confirmed':'y'}, inplace=True)
test.rename(columns={'Dates': 'ds', 'Confirmed':'y'}, inplace=True)

# defining the model

proph = Prophet(growth='logistic', changepoints=['2021-07-21', '2021-07-30', '2021-08-01',
                                                '2021-08-03', '2021-08-10'])

pop = 211463256 # brazil population size
train['cap'] = pop

# train the model
proph.fit(train)

# construction the future predictions
future_dates = proph.make_future_dataframe(periods=200)
future_dates['cap'] = pop
forecast = proph.predict(future_dates)

INFO:prophet:Disabling yearly seasonality. Run prophet with yearly_seasonality=True to override this.
INFO:prophet:Disabling daily seasonality. Run prophet with daily_seasonality=True to override this.

fig = go.Figure()

fig.add_trace(go.Scatter(x=forecast.ds, y=forecast.yhat, name='Predictions'))
#fig.add_trace(go.Scatter(x=test.index, y=test, name='Observed - Test'))
fig.add_trace(go.Scatter(x=train.ds, y=train.y, name='Observed - Train'))
fig.update_layout(title='Predictions of Confirmed Cases in Brazil')
fig.show()

# Mobile mean last 7 days
data['ConfirmedMM'] = data.Confirmed.rolling(7).mean()
data['DeathMM'] = data.Deaths.rolling(7).mean() 

<ipython-input-26-2e40d7a0c8df>:2: SettingWithCopyWarning:


A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy

<ipython-input-26-2e40d7a0c8df>:3: SettingWithCopyWarning:


A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy

data.head(10)

	Country	Dates	Confirmed	Deaths	newcases	ConfirmedMM	DeathMM
15375	Brazil	2020-03-17	321	1	0	NaN	NaN
15654	Brazil	2020-03-18	372	3	51	NaN	NaN
15933	Brazil	2020-03-19	621	6	249	NaN	NaN
16212	Brazil	2020-03-20	793	11	172	NaN	NaN
16491	Brazil	2020-03-21	1021	15	228	NaN	NaN
16770	Brazil	2020-03-22	1546	25	525	NaN	NaN
17049	Brazil	2020-03-23	1924	34	378	NaN	13.571429
17328	Brazil	2020-03-24	2247	46	323	NaN	20.000000
17607	Brazil	2020-03-25	2554	59	307	NaN	28.000000
17886	Brazil	2020-03-26	2985	77	431	NaN	38.142857

plt.plot(data['Deaths'].rolling(7).mean(),label= 'MM Deaths 7 days')
plt.legend(loc='lower right')
plt.title('Mobile Mean of Deaths - 7 days')
plt.show()

plt.plot(data['Confirmed'].rolling(7).mean(),label= 'MM Cases 7 days')
plt.legend(loc='lower right')
plt.title('Mobile Mean of Confirmed Cases - 7 days')
plt.show()